Add application file

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+vosk-model-small-en-us-0.15/** filter=lfs diff=lfs merge=lfs -text
+static/** filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -0,0 +1,25 @@
+# Read the doc: https://huggingface.co/docs/hub/spaces-sdks-docker
+# you will also find guides on how best to write your Dockerfile
+
+FROM python:3.9
+
+# Create a user for security
+RUN useradd -m -u 1000 user
+USER user
+ENV PATH="/home/user/.local/bin:$PATH"
+
+# Set working directory
+WORKDIR /app
+
+# Install Python dependencies
+COPY --chown=user ./requirements.txt requirements.txt
+RUN pip install --no-cache-dir --upgrade -r requirements.txt
+
+# Copy the rest of the code
+COPY --chown=user . /app
+
+# NLTK needs tokenizer data
+RUN python -m nltk.downloader punkt
+
+# Run FastAPI app via Uvicorn on port 7860 (used by Hugging Face Spaces)
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -0,0 +1,90 @@
+from fastapi import FastAPI, Request
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import HTMLResponse
+from fastapi.templating import Jinja2Templates
+from fastapi.staticfiles import StaticFiles
+from fastapi import File, UploadFile
+
+import shutil
+from pydantic import BaseModel
+
+from core import interact_with_user, cache_embeddings
+from voice import transcribe_audio
+
+from json_parser import tree_from_json
+
+from tree import struct_tree
+from time import time
+
+
+# -------- FastAPI setup --------
+app = FastAPI()
+
+# Enable CORS (optional if serving HTML from FastAPI itself)
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],  # In production, replace with specific domains
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+app.mount("/static", StaticFiles(directory="static"), name="static")
+
+# Set up HTML templating
+templates = Jinja2Templates(directory="templates")
+
+@app.get("/", response_class=HTMLResponse)
+async def serve_index(request: Request):
+    return templates.TemplateResponse("index.html", {"request": request})
+
+
+# -------- Load tree and model once --------
+filepath_root = "tree_data"
+filepath_naive = "portfolio_tree_data.json" # Changing this line should change the context completely.
+filepath = f"{filepath_root}/{filepath_naive}"
+
+print("\n--- Loading model and data ---")
+
+s = time()
+tree_file = tree_from_json(filepath)
+e = time()
+print(f"Time taken to load tree from json: {e - s:.2f} seconds")
+
+s = time()
+cache_embeddings(tree_file)
+e = time()
+print(f"Time taken to cache embeddings: {e - s:.2f} seconds")
+
+tree_file.save(filepath[:-5] + ".pkl")
+
+s = time()
+tree_data = struct_tree.load(filepath[:-5] + ".pkl")
+e = time()
+print(f"Time taken to load tree from pkl: {e - s:.2f} seconds")
+
+print("Amber is ready.\n")
+
+
+# -------- API Endpoint --------
+class Query(BaseModel):
+    query: str
+
+@app.post("/ask")
+def ask_amber(query: Query):
+    user_input = query.query
+    response = interact_with_user(tree_data, user_input)
+    return {"response": response}
+
+
+@app.post("/transcribe")
+async def transcribe(file: UploadFile = File(...)):
+    filename = file.filename
+    filepath = f"temp_audio/{filename}"
+
+    with open(filepath, "wb") as buffer:
+        shutil.copyfileobj(file.file, buffer)
+
+    # Call the transcribe function
+    result = transcribe_audio(filepath)
+    return {"transcription": result}

core.py

@@ -0,0 +1,235 @@
+import os
+os.environ['TF_CPP_MIN_LOG_LEVEL'] = '1' # Silence warnings
+
+from sentence_transformers import SentenceTransformer, util
+
+from nltk.tokenize import sent_tokenize
+
+from util import select_random_from_list
+
+from math import exp
+
+all_MiniLM_L12_v2 = SentenceTransformer("all-MiniLM-L12-v2")
+"""
+all-MiniLM-L12-v2 is a sentence embedding model used for tasks involving semantic textual similarity, clustering, semantic search, and information retrieval. They convert a sentence to tensor based on their intent and then matching patterns like cos_sim can be used to compare them to other sentences.
+"""
+
+EXIT_MESSAGE = "Feel free to return any time you want."
+
+CONNECTION_PHRASES = ["just like that", "for the same", "similarly", "similar to the previous", "for that", "for it"]
+CONNECTION_ENCODE = all_MiniLM_L12_v2.encode(CONNECTION_PHRASES,convert_to_tensor=True)
+
+prev_label = ""
+prev_query_data = [] # Stores previous context if queries contain ambiguous content that may map to previous responses
+
+confidence_threshold = 0.35 # Default confidence threshold
+def generate_confidence_threshold(query: str, base=0.6, decay=0.03, min_threshold=0.25)->float:
+    """Generate confidence threshold based on the sentence. Longer sentence lead to lower confidences, so confidence threshold is adjusted based on that.
+    Parameters:
+    1. query: Modify threshold based on this sentence
+    2. base, decay: 0.8*e^(-(decay * no. of words in string))
+    3. min_threshold: Clamp to minimum to avoid much lower confidence values"""
+    global confidence_threshold
+    length = len(query.split())
+    confidence_threshold = max(base * exp(-decay * length), min_threshold)
+
+# The value of each node contains the following data
+# node.value[0] -> intent
+# node.value[1] -> label
+# node.value[2] -> examples
+# node.value[3] -> response
+# node.value[4] -> children
+
+def cache_embeddings(tree, model = all_MiniLM_L12_v2)->None:
+    """Store the encoded examples as part of the tree itself to avoid repetitive computations.
+    Parameters:
+    1. tree: Tree to cache embeddings
+    2. model: Which model to use to encode (Default: Global model all_MiniLM_L12_v2)"""
+
+    def _cache_node_embeddings(n):
+        if isinstance(n.value, tuple) and len(n.value) >= 2:
+            examples = n.value[2]
+            n.embedding_cache = model.encode(examples, convert_to_tensor=True)
+        for child in n.children:
+            _cache_node_embeddings(child)
+    _cache_node_embeddings(tree.root)
+
+CLEAR_MESSAGES = ["delete", "delete context", "delete history", "clear", "clear context", "clear history", "reset", "reset context", "reset chat", "forget", "forget all"]
+CLEAR_MESSAGES_ENCODE = all_MiniLM_L12_v2.encode(CLEAR_MESSAGES,convert_to_tensor=True)
+
+SECOND_PERSON_MENTIONS = ["you", "your", "yours", "yourself", "y'all", "y'all's", "y'all'self", "you're", "your'e""u", "ur", "urs", "urself"]
+def get_user_query(message="", model = all_MiniLM_L12_v2)->str:
+    """Separate function to get input from user.
+    Parameters:
+    1. message: Show message to user before recieving input (Default: empty)
+    2. model: Which model to use to encode (Default: Global model all_MiniLM_L12_v2)"""
+
+    query = input(message).lower().strip()
+    while query == "":
+        query = input(message).lower().strip()
+    
+    query = query.replace("  "," ") # Remove double spaces
+    for spm in SECOND_PERSON_MENTIONS: # Remove second person mentions
+        query = query.replace(spm,"Amber AI")  # Replace with bot name
+    generate_confidence_threshold(query)
+    query_encode = model.encode(query, convert_to_tensor=True)
+    clear_intent = util.cos_sim(query_encode,CLEAR_MESSAGES_ENCODE).max().item()
+    if clear_intent > confidence_threshold:
+        return None
+    return query
+
+def _calculate_single_level(user_embed,predicted_intent):
+    """Calculate predictions for the children of a single node. Each node contains a list of nodes as its children.
+    Parameters:
+    1. user_embed: User query converted to tensor
+    2. predicted_intent: Calculate for children of this node"""
+
+    categories = predicted_intent.children # List of node objects
+    predicted_intent = None
+    high_intent = 0
+    for category in categories:
+        if category.embedding_cache is None:
+            raise ValueError("Embedding cache missing. Call cache_embeddings() on the tree first")
+        score = util.cos_sim(user_embed, category.embedding_cache).max().item()
+
+        if score > high_intent:
+            high_intent = score
+            predicted_intent = category # Node object
+    return (predicted_intent,high_intent) # Returns the child node with the highest prediction confidence and the confidence value
+
+def _store_prev_data(predicted_intent):
+    """Store the previous computed data path.
+    Parameters:
+    1. predicted_intent: Store previous data w.r.t this node"""
+    # Mutating global prev_query_data
+    prev_query_data.clear()
+    prev_context_treenode = predicted_intent
+    while prev_context_treenode.parent: # Stop at tree root
+        prev_query_data.append(prev_context_treenode)
+        prev_context_treenode = prev_context_treenode.parent
+
+def h_pass(tree, user_embed, predicted_intent = None)->tuple:
+    """Use the model to pass through the tree to compare it with the user query in a hierarchical manner and return an output.
+    Parameters:
+    1. tree: Which tree to pass through hierarchically
+    2. user_embed: User input converted to tensor
+    3. predicted_intent: Where to start the pass from (Default: Root of the tree)"""
+    global prev_label
+    predicted_intent = tree.root if predicted_intent == None else predicted_intent
+    predicted_intent_parent = None
+    high_intent = 0
+    passed_once = False
+    pass_through_intent = {}
+    while predicted_intent.children: # If the node has children, check for the child with the highest confidence value
+        predicted_intent_parent = predicted_intent
+        predicted_intent, high_intent = _calculate_single_level(user_embed,predicted_intent)
+        pass_through_intent[predicted_intent] = high_intent # Store the confidence value of the current node
+        if passed_once: # If the data didn't pass even once, then don't store it
+            _store_prev_data(predicted_intent_parent) # Storing previous data w.r.t parent node as context is changed from current node
+        if high_intent < confidence_threshold: # If highest confidence value is still too low, stop.
+            prev_label = predicted_intent_parent.value[1]
+            return (predicted_intent, predicted_intent_parent, high_intent, passed_once, False, pass_through_intent) # If the confidence value is low, stop   
+        passed_once = True
+    
+    _store_prev_data(predicted_intent)
+    prev_label = predicted_intent.value[1]
+    return (predicted_intent, predicted_intent_parent, high_intent, passed_once, True, pass_through_intent)
+
+def query_pass(tree, user_input, model=all_MiniLM_L12_v2)->list:
+    """Separate multiple queries into separate single ones, analyze relation between them if any, and process them to give an output while storing incomplete query outputs in non-leaf list, which contains the current level of context.
+    Parameters:
+    1. tree: Which tree to pass through hierarchically
+    2. user_input: User input that may contain one or more queries as a string
+    3. model: Which model to use to encode (Default: Global model all_MiniLM_L12_v2)"""
+
+    queries = sent_tokenize(user_input)
+    user_embeddings = [model.encode(query,convert_to_tensor=True) for query in queries]
+    result = []
+    label = prev_label
+
+    for i in range(len(queries)):
+        generate_confidence_threshold(queries[i])
+        pass_value = (None, None, 0, False, False, None)
+        # pass_value[0] -> current predicted intention (node)
+        # pass_value[1] -> parent node of current predicted intention
+        # pass_value[2] -> confidence level
+        # pass_value[3] -> has the query passed through the model atleast once?
+        # pass_value[4] -> has the query reached a leaf node?
+        # pass_value[5] -> confidence values of traversal for query [DEBUGGING PURPOSES]
+
+        # Acquiring data from previous query if the query has words matching with connecting phrases
+        conn_sim = util.cos_sim(user_embeddings[i], CONNECTION_ENCODE).max().item()
+        if conn_sim > confidence_threshold:
+            queries[i] = queries[i] + label
+            user_embeddings[i] = model.encode(queries[i], convert_to_tensor=True)
+
+        # Pass values through the root node and the nodes that have the current context
+        pass_value_root = h_pass(tree,user_embeddings[i]) # Passing through root node
+        pass_value_nonleaf = [h_pass(tree,user_embeddings[i],j) for j in prev_query_data] # Passing through nodes that have current context
+        all_nodes = [pass_value_root] + pass_value_nonleaf # List of all nodes that have been passed through
+        pass_value = max(all_nodes, key=lambda x: x[2]) # Maximum confidence node for available context. Root is always a context.
+        print(f"Query reach confidence: {[i[5] for i in all_nodes]}") # DEBUGGING PURPOSES
+
+        if pass_value[3]: # If the query has passed at least once, ask for data and store current result
+            if not pass_value[4]: # If pass has not reached a leaf node, then ask for more data from the user and keep parent context
+                label = pass_value[1].value[1]
+                result.append(f"{pass_value[1].value[3]}")
+                # continue
+
+            else: # Query has reached a leaf node
+                label = pass_value[0].value[1]
+                result.append(pass_value[0].value[3])
+                # continue
+
+        else: # Query has not passed even once. Check if it works when previous context is available
+            for parent_context in prev_query_data:
+                pass_value_context = h_pass(tree, user_embeddings[i], parent_context)
+                if pass_value_context[3]: # Check if it has passed at least once
+                    # If it has passed, then the query is valid
+                    if not pass_value_context[4]: # If pass has not reached a leaf node, then ask for more data from the user and keep parent context
+                        label = pass_value_context[1].value[1]
+                        result.append(f"What are you looking for in {pass_value_context[1].value[0]}? {pass_value_context[1].value[3]}")
+                    else:
+                        label = pass_value_context[0].value[1]
+                        result.append(pass_value_context[0].value[3])
+                    break # The else block won't be executed if code reaches here
+            else: # The else statement of a for loop will execute only if the loop completes, and won't execute when broken by "break"
+                result.append(f"I don't quite understand what you are trying to ask by \"{queries[i]}\"")
+                # continue
+    # End of "for" loop processing queries
+
+    # Finally, return result. A list of responses same as the length of queries.
+    return result
+
+def process_user_query(query: str, model = all_MiniLM_L12_v2)->str:
+    """Separate function to get input from user.
+    Parameters:
+    1. message: Show message to user before recieving input (Default: empty)
+    2. model: Which model to use to encode (Default: Global model all_MiniLM_L12_v2)"""
+
+    query = query.lower().strip()
+    generate_confidence_threshold(query)
+    query_encode = model.encode(query, convert_to_tensor=True)
+    clear_intent = util.cos_sim(query_encode,CLEAR_MESSAGES_ENCODE).max().item()
+    if clear_intent > confidence_threshold:
+        return None
+    return query
+
+def interact_with_user(tree_data, user_input: str) -> str:
+    """Handles a single user query and returns a response string."""
+    user_input = process_user_query(user_input)
+    all_results = []
+    if user_input:  # If not empty or command
+        results = query_pass(tree_data, user_input)
+        for result in results:
+            # return f"{select_random_from_list(result)}\nContext window: {prev_query_data}"
+            all_results.append(f"{select_random_from_list(result)}")
+        print(f"Previous query data: {prev_query_data}")
+        return all_results
+    else:
+        # Mutating global variables: Clearing context and recent history on command
+        prev_query_data.clear()
+        print(f"Previous query data: {prev_query_data}")
+        return ["Cleared previous context"]
+    
+    

documentation/AmberAI_concept.txt

@@ -0,0 +1,149 @@
+# 📄 **Composable Answer Engine: A Deterministic, Low-Compute Alternative to Generative Chatbots**
+
+## ✦ Overview
+
+The **Composable Answer Engine** is a novel approach to building intelligent, language-aware question-answering systems. Unlike conventional chatbots (which are rigid and brittle) or large language models (which are resource-intensive and prone to hallucination), this system offers a **middle path**: it provides natural, flexible, and contextually relevant responses by **composing pre-approved answer fragments** rather than generating free-form language.
+
+The system is designed to be **safe, efficient, and domain-controlled**, making it ideal for business applications where **accuracy, auditability, and trust** are more important than creative generation.
+
+---
+
+## ✦ Problem Statement
+
+Existing approaches to automated customer or user interaction fall into two categories:
+
+| Approach | Description | Limitations |
+|---------|-------------|-------------|
+| **Rule-based chatbots** | Pre-scripted flows, keyword matching | Inflexible, brittle, poor handling of paraphrasing |
+| **LLM-powered assistants** | Use transformers to generate fluent answers | Require high compute, risk hallucination, hard to control, privacy/legal concerns |
+
+**Challenge:**  
+There is no widely adopted approach that combines:
+- Language understanding
+- Deterministic, auditable answers
+- Low computational overhead
+- Modular knowledge expansion
+
+---
+
+## ✦ Proposed Solution
+
+The **Composable Answer Engine (CAE)** answers natural language questions by **matching user input to semantically similar phrases** and **stitching together pre-written, approved text segments**.
+
+It avoids neural generation entirely and instead uses **retrieval + controlled composition** to ensure:
+- No hallucinations
+- Full traceability
+- Cost-effective deployment
+
+---
+
+## ✦ System Architecture
+
+```plaintext
++------------------------+
+|   User Query (Text)    |
++-----------+------------+
+            |
+            v
++------------------------+
+| Embed with Sentence    |
+| Transformer (e.g.,     |
+| MiniLM or MPNet)       |
++-----------+------------+
+            |
+            v
++------------------------+
+| Semantic Similarity    |
+| Search (e.g., FAISS,   |
+| cosine sim over corpus)|
++-----------+------------+
+            |
+            v
++------------------------+
+| Phrase Assembly Module |
+| - Match answer blocks  |
+| - Apply rules if needed|
++-----------+------------+
+            |
+            v
++------------------------+
+|    Output Answer       |
++------------------------+
+```
+
+---
+
+## ✦ Key Components
+
+### 1. **Sentence Embeddings**
+- Uses small pre-trained models (e.g., `all-MiniLM-L6-v2`) from [Sentence Transformers](https://www.sbert.net/) to embed queries and stored phrases into a shared semantic space.
+
+### 2. **Answer Corpus**
+- Structured markdown, YAML, or database entries containing:
+  - `questions:` → example paraphrases
+  - `answer:` → approved response text
+- Easily editable by non-technical staff.
+
+### 3. **Similarity Engine**
+- Uses cosine similarity or vector indexing (FAISS) to retrieve the closest matching content block(s).
+
+### 4. **Answer Composer**
+- Returns the most appropriate answer verbatim or assembles one using adjacent text blocks (if rule-based composition is enabled).
+
+### 5. **Fallback Handler**
+- If confidence is low, the system can:
+  - Prompt the user to rephrase
+  - Escalate to a human
+  - Log the query for admin review and knowledge base expansion
+
+---
+
+## ✦ Benefits
+
+| Feature | Advantage |
+|--------|-----------|
+| 💻 **Low compute** | Runs on CPU; no GPUs or cloud inference needed |
+| 🛑 **No hallucination** | Outputs only pre-approved content |
+| 🔐 **High privacy** | No sensitive data leaves the device |
+| ✏️ **Easily editable** | Add new knowledge via flat files (Markdown/YAML/CSV) |
+| 🧠 **Semantic understanding** | Handles paraphrasing and natural queries |
+| 📑 **Explainable** | Every output is traceable to exact input fragments |
+| 🔄 **Extendable** | Modular design allows domain-specific packs |
+
+---
+
+## ✦ Limitations
+
+| Limitation | Description |
+|------------|-------------|
+| ❌ No generative creativity | Cannot form novel phrasing outside stored fragments |
+| ⚖️ Depends on knowledge base quality | Poorly structured content = poor results |
+| 🧩 Answer composition complexity | Combining fragments can sound robotic if not carefully curated |
+| 📚 No "reasoning" | Cannot synthesize new conclusions or multi-hop logic |
+
+---
+
+## ✦ Ideal Use Cases
+
+- Internal company assistants (HR, IT helpdesk)
+- Customer service bots with strict compliance needs
+- Healthcare, finance, law domains requiring controlled outputs
+- Offline-first applications (e.g., field tools, kiosks)
+- Educational tools where factual consistency is vital
+
+---
+
+## ✦ Future Work
+
+- **Lightweight personalization** using user preferences stored locally
+- **Admin dashboard** to monitor unanswered queries and expand corpus
+- **Domain-specific packs** (e.g., `legal_pack.md`, `finance_pack.md`)
+- **Optional smoothing layer** (rule-based grammar post-processing)
+
+---
+
+## ✦ Summary
+
+The **Composable Answer Engine** is a safe, practical, and explainable alternative to current chatbot paradigms. It allows organizations to deliver intelligent, responsive user interaction **without the risks, costs, or unpredictability** of full LLM-based solutions.
+
+Its hybrid approach — *semantic understanding + controlled response synthesis* — fills a major gap in the current AI ecosystem, especially in domains that demand trust, traceability, and low resource usage.

documentation/AmberAI_effect.txt

@@ -0,0 +1,7 @@
+Designed a hybrid retrieval-based QA system using MiniLM for low-cost intent matching and FAISS for fast embedding lookup, enabling sub-100ms query response time.
+
+Built a deterministic fallback logic engine to route user queries to curated, trusted answer sets without hallucination.
+
+Optimized embedding pipeline for domain-specific intents (e.g., job queries, internal policies), achieving >90% hit accuracy across paraphrased inputs.
+
+Engineered with deployment in mind: containerized, minimal memory footprint (<300MB), and LLM-free.

documentation/AmberAI_examples.txt

@@ -0,0 +1,79 @@
+🔹 1. Course Offerings
+Chatbot:
+Query: Courses
+Answer: AI, Data Science
+Prompt: Click on the course you want to apply for or click back
+
+LLM:
+Query: What courses does this company offer?
+Answer: Our company offers courses on AI and Data Science. Would you like to apply for one?
+Prompt: None, the user enters it.
+
+StitchAI:
+Query: What courses does this company offer?
+Answer: AI, Data Science [Not generated]
+Prompt: None, the user enters it.
+
+🔹 2. Office Hours
+Chatbot:
+Query: Office hours
+Answer: 9AM – 5PM, Mon to Fri
+Prompt: Press 1 to view branches, press 2 to go back
+
+LLM:
+Query: When is your office open during the week?
+Answer: Our office is open from 9AM to 5PM, Monday to Friday.
+Prompt: None
+
+StitchAI:
+Query: When is your office open during the week?
+Answer: 9AM – 5PM, Mon to Fri [Not generated]
+Prompt: None
+
+🔹 3. Contact Information
+Chatbot:
+Query: Contact
+Answer: support@example.com
+Prompt: Click to send email or go back
+
+LLM:
+Query: How can I contact your support team?
+Answer: You can reach our support team at support@example.com.
+Prompt: None
+
+StitchAI:
+Query: How can I contact your support team?
+Answer: support@example.com [Not generated]
+Prompt: None
+
+🔹 4. Refund Policy
+Chatbot:
+Query: Refunds
+Answer: Refunds only within 7 days
+Prompt: Click here to request refund
+
+LLM:
+Query: What's your company's refund policy?
+Answer: Our company provides refunds only within 7 days of purchase.
+Prompt: None
+
+StitchAI:
+Query: What's your company's refund policy?
+Answer: Refunds only within 7 days [Not generated]
+Prompt: None
+
+🔹 5. Job Openings
+Chatbot:
+Query: Careers
+Answer: No openings currently
+Prompt: Go back
+
+LLM:
+Query: Are there any current job openings at your company?
+Answer: Currently, we do not have any open positions. Please check back later.
+Prompt: None
+
+StitchAI:
+Query: Are there any current job openings at your company?
+Answer: No openings currently [Not generated]
+Prompt: None

json_parser.py

@@ -0,0 +1,46 @@
+import json
+import os
+from tree import struct_tree 
+
+SECOND_PERSON_MENTIONS = ["you", "your", "yours", "yourself", "y'all", "y'all's", "y'all'self", "you're", "your'e""u", "ur", "urs", "urself"]
+
+def tree_from_json(filepath):
+    """
+    Load a tree from a structured JSON file.
+    Root node name is taken from the filename (without extension).
+    Each node must have: intent, examples, response, and optionally children.
+    """
+    with open(filepath, 'r') as f:
+        data = json.load(f)
+
+    filename_intent = os.path.splitext(os.path.basename(filepath))[0]
+    root_value = (filename_intent, ["Everything"], data["response"])
+    tree = struct_tree(root_value)
+
+    def add_children(parent_node, children_data):
+        for child in children_data:
+            # Validation
+            if not all(k in child for k in ("intent", "label", "examples", "response")):
+                raise ValueError(f"Missing required fields in node: {child}")
+            
+            intent = child["intent"]
+            label = child["label"]
+
+            examples = child["examples"]
+            # Replace second-person mentions with "Amber AI" to match context
+            # for i in range(len(examples)):
+            #     for spm in SECOND_PERSON_MENTIONS:
+            #         examples[i] = examples[i].replace(spm, "Amber AI")
+
+            response = child["response"]
+
+            value = (intent, label, examples, response)
+            new_node = tree.add_node(parent_node, value)
+
+            if "children" in child:
+                add_children(new_node, child["children"])
+
+    if "children" in data:
+        add_children(tree.root, data["children"])
+    
+    return tree

requirements.txt

@@ -0,0 +1,8 @@
+fastapi
+uvicorn
+sentence-transformers
+nltk
+pydantic
+vosk
+jinja2
+python-multipart

static/opensans.ttf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6530480277da62efde047eb26e78a7e532d1cfaeec91603e68d63876b9669f0d
+size 130832

static/script.js

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:748cc42518a80b222f2ddc36f4af2c9e6819a95e810642b234eb0908205d951f
+size 4642

static/styles.css

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47832d78b799fd1f20603a7601510ef0bd111d90b05e6f6b014feaa3b9ff155f
+size 4206

templates/index.html

@@ -0,0 +1,221 @@
+<html>
+
+<head>
+  <title>Amber</title>
+  <link rel="stylesheet" href="/static/styles.css" />
+  <link rel="icon" href="/static/favicon.ico" type="image/x-icon">
+</head>
+
+<body>
+  <div class="box" id="interactive">
+
+    <div class="chatc" id="chatc">
+      <input type="text" placeholder="Ask Amber" id="chat" />
+      <img src="/static/mic.png" id="mic">
+      <img src="/static/refresh.png" id="ref">
+
+      <!-- Chat section -->
+      <p>
+        <font style="color: #f23000; font-weight: bold;">Amber</font> <br>
+        Hi. I am Amber. Ask me anything.
+      </p>
+      <!-- End of chat section -->
+    </div>
+  </div>
+  <div class="side">
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Introduction</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Who am I?</font> <br>
+      Hi. I am Dwarakesh. Aspiring AI/ML engineer with a deep interest in how machines think and how they render pixels
+      to
+      the screen.
+    </p>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Tech interests</font> <br>
+      I love to explore how machines learn and how GPUs render—juggling AI/ML projects and Vulkan code to understand
+      pixels and neural nets. Curiosity-driven, stubbornly hands-on.
+    </p>
+    <br>
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Education</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">School and College</font> <br>
+      I studied in PSBB Millennium School, Coimbatore up until grade 8, Suguna Pip School, Coimbatore until grade 12,
+      and
+      pursuing my B.Tech in Computer Science and Engineering at Amrita Vishwa Vidyapeetham, Coimbatore.
+    </p>
+    <br>
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Skills and expertise</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Programming languages</font> <br>
+      &#8226; Python<br>
+      &#8226; Rust<br>
+      &#8226; C++<br>
+      &#8226; HTML<br>
+      &#8226; CSS<br>
+      &#8226; JavaScript<br>
+      &#8226; Java<br>
+      &#8226; Kotlin<br>
+      &#8226; Golang<br>
+      &#8226; Haskell<br>
+    </p>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Tech stack</font> <br>
+      &#8226; Vulkan<br>
+      &#8226; OpenGL<br>
+      &#8226; PyTorch<br>
+      &#8226; Tensorflow<br>
+      &#8226; OpenCV<br>
+      &#8226; Django<br>
+      &#8226; FastAPI<br>
+      &#8226; Tokio<br>
+      &#8226; Rayon<br>
+      &#8226; WebGPU<br>
+      &#8226; CUDA<br>
+    </p>
+  </div>
+  <div class="side">
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Projects</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Amber</font> <br>
+      A chatbot fine-tuned on the MiniLM-L12-v2 model. It maps human language to structured answers using a hierarchical
+      traversal of its knowledge base. Designed to adapt context or update knowledge on the fly.
+      <br>
+      <a href="github.com"> &#8599; View on Github </a>
+    </p>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Arx Net</font> <br>
+      An open-source graph visualization tool powered by WebGPU. Built for real-time interaction with large graphs—ideal
+      for networks in biology, social media, or systems modeling.
+      <br>
+      <a href="github.com"> &#8599; View on Github </a>
+    </p>
+    <br>
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Work experience and competitions</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Internships, Jobs, and Competitions</font> <br>
+      Participated in 3 hackathons. <br>
+      No internships or work experience (yet)
+    </p>
+    <br>
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Contact</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold; font-size: 20px;">Contact link</font> <br>
+      &#8226; Email: <a href="https://www.github.com/Dwarakesh-V"> dwarakesh.2005.4@gmail.com </a> <br>
+      &#8226; Github: <a href="https://www.github.com/Dwarakesh-V"> https://www.github.com/Dwarakesh-V </a> <br>
+      &#8226; LinkedIn: <a href="https://www.linkedin.com/in/vdwarakesh/"> https://www.linkedin.com/in/vdwarakesh/
+      </a>
+      <br>
+    </p>
+  </div>
+  <div class="statics" id="statics">
+    <img src="/static/refresh.png" id="ref2">
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Introduction</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Who am I?</font> <br>
+      Hi. I am Dwarakesh. Aspiring AI/ML engineer with a deep interest in how machines think and how they render pixels
+      to
+      the screen.
+    </p>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Tech interests</font> <br>
+      I love to explore how machines learn and how GPUs render—juggling AI/ML projects and Vulkan code to understand
+      pixels and neural nets. Curiosity-driven, stubbornly hands-on.
+    </p>
+    <br>
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Education</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">School and College</font> <br>
+      I studied in PSBB Millennium School, Coimbatore up until grade 8, Suguna Pip School, Coimbatore until grade 12,
+      and
+      pursuing my B.Tech in Computer Science and Engineering at Amrita Vishwa Vidyapeetham, Coimbatore.
+    </p>
+    <br>
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Skills and expertise</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Programming languages</font> <br>
+      &#8226; Python<br>
+      &#8226; Rust<br>
+      &#8226; C++<br>
+      &#8226; HTML<br>
+      &#8226; CSS<br>
+      &#8226; JavaScript<br>
+      &#8226; Java<br>
+      &#8226; Kotlin<br>
+      &#8226; Golang<br>
+      &#8226; Haskell<br>
+    </p>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Tech stack</font> <br>
+      &#8226; Vulkan<br>
+      &#8226; OpenGL<br>
+      &#8226; PyTorch<br>
+      &#8226; Tensorflow<br>
+      &#8226; OpenCV<br>
+      &#8226; Django<br>
+      &#8226; FastAPI<br>
+      &#8226; Tokio<br>
+      &#8226; Rayon<br>
+      &#8226; WebGPU<br>
+      &#8226; CUDA<br>
+    </p>
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Projects</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Amber</font> <br>
+      A chatbot fine-tuned on the MiniLM-L12-v2 model. It maps human language to structured answers using a hierarchical
+      traversal of its knowledge base. Designed to adapt context or update knowledge on the fly.
+      <br>
+      <a href="github.com"> &#8599; View on Github </a>
+    </p>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Arx Net</font> <br>
+      An open-source graph visualization tool powered by WebGPU. Built for real-time interaction with large graphs—ideal
+      for networks in biology, social media, or systems modeling.
+      <br>
+      <a href="github.com"> &#8599; View on Github </a>
+    </p>
+    <br>
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Work experience and competitions</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Internships, Jobs, and Competitions</font> <br>
+      Participated in 3 hackathons. <br>
+      No internships or work experience (yet)
+    </p>
+    <br>
+    <center>
+      <font style="color: #f23000; font-weight: bold;">Contact</font>
+    </center>
+    <p>
+      <font style="color: #0087f2; font-weight: bold;">Contact link</font> <br>
+      &#8226; Email: <a href="https://www.github.com/Dwarakesh-V"> dwarakesh.2005.4@gmail.com </a> <br>
+      &#8226; Github: <a href="https://www.github.com/Dwarakesh-V"> https://www.github.com/Dwarakesh-V </a> <br>
+      &#8226; LinkedIn: <a href="https://www.linkedin.com/in/vdwarakesh/"> https://www.linkedin.com/in/vdwarakesh/
+      </a>
+      <br>
+    </p>
+  </div>
+  <script src="/static/script.js"></script>
+</body>
+
+</html>

tree.py

@@ -0,0 +1,78 @@
+import pickle
+
+class node:
+    """Individual node class of each node of tree"""
+    def __init__(self, value=None):
+        self.value = value
+        self.parent = None
+        self.children = []
+        self.embedding_cache = None  # Will hold precomputed embeddings if available
+    
+    def __repr__(self):
+        return str(self.value[0]) if isinstance(self.value, tuple) else str(self.value)
+
+class struct_tree:
+    def __init__(self,value): # Root is mandatory
+        self.root = node(value)
+    
+    def add_node(self, parent_node: node, value): # Add a node
+        """Add node to the tree."""
+        new_node = node(value)
+        new_node.parent = parent_node
+        parent_node.children.append(new_node)
+        return new_node
+    
+    def visualize(self, show_labels=True):
+        """Display the tree in a graph-like format."""        
+        def _print_node(current_node, prefix="", is_last=True):
+            # Print current node
+            branch = "└── " if is_last else "├── "
+            print(f"{prefix}{branch}{current_node.value if show_labels else 'o'}")
+            
+            # Prepare prefix for children
+            extension = "    " if is_last else "│   "
+            new_prefix = prefix + extension
+            
+            # Print children
+            child_count = len(current_node.children)
+            for i, child in enumerate(current_node.children):
+                is_last_child = i == child_count - 1
+                _print_node(child, new_prefix, is_last_child)
+        
+        # Start recursive printing from root
+        print(f"{self.root.value if show_labels else 'o'}")
+        child_count = len(self.root.children)
+        for i, child in enumerate(self.root.children):
+            is_last_child = i == child_count - 1
+            _print_node(child, "", is_last_child)
+    
+    def save(self, filepath: str):
+        """Save the entire tree structure to disk with optional compression."""
+        with open(filepath, 'wb') as f:
+            pickle.dump(self, f)
+
+    @staticmethod # This method is called directly on the class rather than an instance of it
+    def load(filepath: str):
+        """Load the tree structure from disk."""
+        with open(filepath, 'rb') as f:
+            return pickle.load(f)
+
+if __name__ == "__main__": # Execute this only in this file
+    # Create a tree with root value "A"
+    tree = struct_tree("A")
+    
+    # Add some nodes
+    b_node = tree.add_node(tree.root, "B")
+    c_node = tree.add_node(tree.root, "C")
+    d_node = tree.add_node(tree.root, "D")
+    
+    # Add children to B
+    tree.add_node(b_node, "B1")
+    b2_node = tree.add_node(b_node, "B2")
+    tree.add_node(b2_node, "B2.1")
+    
+    # Add children to C
+    tree.add_node(c_node, "C1")
+    
+    # Visualize the tree
+    tree.visualize(show_labels=False)

tree_data/company_tree_data.json

@@ -0,0 +1,40 @@
+{
+  "intent": "company",
+  "label":"company",
+  "examples": ["Everything"],
+  "response": "Ask me anything related to the company!",
+  "children": [
+    {
+      "intent": "job opportunities",
+      "label": "ai and data science",
+      "examples": ["Are there jobs available?", "Looking for work", "Any openings?"],
+      "response": "We currently have job openings in multiple domains like Data Science and AI/ML.",
+      "children": [
+        {
+          "intent": "data science",
+          "label": "data science",
+          "examples": ["data science", "Interested in data science"],
+          "response": "We have open positions in Data Science. Visit careers/company.com/data-science."
+        },
+        {
+          "intent": "aiml",
+          "label": "ai",
+          "examples": ["machine learning", "Interested in machine learning and artificial intelligence", "AI", "ML", "AI/ML"],
+          "response": "We're hiring in AI/ML. See careers/company.com/aiml."
+        }
+      ]
+    },
+    {
+      "intent": "courses",
+      "label": "ai and data science",
+      "examples": ["What courses do you have?", "Available classes?", "Do you offer courses?", "Are there any courses?"],
+      "response": "We offer courses in AI, Data Science, and Software Engineering."
+    },
+    {
+      "intent": "contact",
+      "label": "",
+      "examples": ["How can I reach you?", "Contact info please", "Can I call you?", "Can I speak to your", "Can I colab with your", "Manager contact"],
+      "response": "Reach us via email at info@company.com or call 123-456-7890."
+    }
+  ]
+}

tree_data/company_tree_data.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e2a4e4987926874c6f498b65896017c8ab64c5d3bef67e9438084ae71f1c922
+size 35329

tree_data/portfolio_tree_data.json

@@ -0,0 +1,951 @@
+{
+  "intent": "Amber",
+  "label": "Amber",
+  "examples": [],
+  "response": "Hello, I'm Amber. Ask me anything!",
+  "children": [
+    {
+      "intent": "greetings",
+      "label": "greetings",
+      "examples": [
+        "Hi",
+        "Hi there",
+        "Hello",
+        "Hey",
+        "Heya",
+        "Howdy",
+        "Yo",
+        "Good to see you",
+        "Welcome",
+        "Glad you're here",
+        "Nice to have you",
+        "Sup",
+        "Hiya"
+      ],
+      "response": [
+        "Hey there",
+        "Hello",
+        "Hi",
+        "Hey",
+        "Howdy",
+        "Greetings",
+        "Welcome"
+      ]
+    },
+    {
+      "intent": "goodbyes",
+      "label": "goodbyes",
+      "examples": [
+        "Bye!",
+        "Goodbye",
+        "See ya",
+        "Cya",
+        "Catch you later!",
+        "Later!",
+        "Take care",
+        "Peace out",
+        "Talk soon",
+        "See you around",
+        "Have a good one",
+        "See ya later",
+        "I'm out",
+        "Adios"
+      ],
+      "response": [
+        "Catch you later",
+        "See you later",
+        "Goodbye",
+        "Take care",
+        "Bye",
+        "See ya"
+      ]
+    },
+    {
+      "intent": "thanks",
+      "label": "thanks",
+      "examples": [
+        "Thanks",
+        "Thank you",
+        "Much appreciated",
+        "Thanks a lot",
+        "Really appreciate it",
+        "Thanks so much",
+        "Big thanks",
+        "Appreciate it"
+      ],
+      "response": [
+        "Anytime",
+        "Happy to help",
+        "You're welcome",
+        "No problem",
+        "Glad I could help",
+        "Anytime"
+      ]
+    },
+    {
+      "intent": "about",
+      "label": "about",
+      "examples": [
+        "Tell me about yourself",
+        "About you",
+        "Tell me something about you",
+        "Can you tell me about yourself?",
+        "About yourself"
+      ],
+      "response": "I am a student exploring AI/ML and GPU rendering, often diving into Vulkan/low-level graphics. Curiosity-driven projects blend learning with stubborn tinkering."
+    },
+    {
+      "intent": "who are you",
+      "label": "who are you",
+      "examples": [
+        "Who are you?",
+        "What's your name?",
+        "And you are?",
+        "What are you called?",
+        "Mind telling me your name?",
+        "Can I ask your name?",
+        "Who am I talking to?",
+        "What do people call you?",
+        "What are you?",
+        "What am I talking to?"
+      ],
+      "response": [
+        "I am Amber. And I am representing a part of my creator - Dwarakesh.",
+        "I am Amber, a chatbot created by and represents Dwarakesh."
+      ]
+    },
+    {
+      "intent": "what are you doing",
+      "label": "what are you doing",
+      "examples": [
+        "What are you doing?",
+        "What are you up to?"
+      ],
+      "response": [
+        "Chatting with you! Ask me anything related to me or my creator.",
+        "I am here to chat with you! Ask me anything related to me or my creator."
+      ]
+    },
+    {
+      "intent": "where are you from",
+      "label": "where are you from",
+      "examples": [
+        "Where are you from?",
+        "Where do you come from?",
+        "What's your origin?",
+        "Where were you made?",
+        "Where do you live?",
+        "Which place are you based in?",
+        "Where were you born?",
+        "What place are you from?"
+      ],
+      "response": [
+        "I am from Coimbatore. A city in Tamil Nadu, India."
+      ],
+      "children": [
+        {
+          "intent": "country",
+          "label": "country",
+          "examples": [
+            "Which country are you from?",
+            "What country are you in?",
+            "What is your country?",
+            "What country do you belong to?"
+          ],
+          "response": "I am from India, a country in South Asia."
+        },
+        {
+          "intent": "state",
+          "label": "state",
+          "examples": [
+            "Which state are you from?",
+            "What state are you in?",
+            "What is your state?",
+            "What state do you belong to?"
+          ],
+          "response": "I am from Tamil Nadu, a state in South India."
+        },
+        {
+          "intent": "city",
+          "label": "city",
+          "examples": [
+            "Which city are you from?",
+            "What city are you in?",
+            "What is your city?",
+            "What city do you belong to?"
+          ],
+          "response": "I am from Coimbatore, located in Tamil Nadu."
+        }
+      ]
+    },
+    {
+      "intent": "where are you from server",
+      "label": "where are you from server",
+      "examples": [
+        "Server",
+        "Where are you hosted?",
+        "Where do you run?",
+        "Source code",
+        "Where can I see your code?",
+        "Where are you deployed?"
+      ],
+      "response": "I run on Hugging face spaces, and my source code is available on Github under the MIT license."
+    },
+    {
+      "intent": "how are you",
+      "label": "how are you",
+      "examples": [
+        "How are you?",
+        "How's it going?",
+        "How are things?",
+        "You doing okay?",
+        "How have you been?",
+        "Everything good?",
+        "All good with you?"
+      ],
+      "response": [
+        "I'm doing just fine. Hope you are too.",
+        "I'm doing great. Hope that's the case with you too.",
+        "I'm doing well, thanks for asking. Hope you're doing well too."
+      ]
+    },
+    {
+      "intent": "education",
+      "label": "education",
+      "examples": [
+        "Where did you study?",
+        "Education stats",
+        "School",
+        "College",
+        "Which school did you study in?",
+        "Which college did you study in?",
+        "Where are you studying?"
+      ],
+      "response": "I studied at PSBB Millennium till 8th and Suguna Pip School till 12th, both in Coimbatore. I'm currently doing my undergrad at Amrita Vishwa Vidyapeetham, Coimbatore.",
+      "children": [
+        {
+          "intent": "school",
+          "label": "school",
+          "examples": [
+            "School",
+            "Which school did you study in?",
+            "Where was your early education?"
+          ],
+          "response": "I studied at PSBB Millennium till 8th and Suguna Pip School till 12th, both in Coimbatore. As for kindergarten, I was in Akshara fun School, Coimbatore"
+        },
+        {
+          "intent": "college",
+          "label": "college",
+          "examples": [
+            "College",
+            "Which college did you study in?",
+            "Where did you do your UG?",
+            "Where did you complete your undergraduate degree?",
+            "Are you pursuing your undergraduate degree?"
+          ],
+          "response": "I'm currently pursuing my undergraduate degree for B.TECH Computer Science in Amrita Vishwa Vidyapeetham, Coimbatore"
+        }
+      ]
+    },
+    {
+      "intent": "exam fail",
+      "label": "exam fail",
+      "examples": [
+        "Have you failed an exam?",
+        "Did you ever mess up an exam?",
+        "Have you failed before?",
+        "Ever gotten something totally wrong?",
+        "Have you had a bad result?"
+      ],
+      "response": "I have failed in my JEE Advanced exam. My score was 48, while the cutoff was 55, out of 360."
+    },
+    {
+      "intent": "arrears",
+      "label": "arrears",
+      "examples": [
+        "Arrears",
+        "Do you have any arrears?",
+        "Any pending subjects to clear?",
+        "Any backlogs?"
+      ],
+      "response": [
+        "Nope! I'm all clear - never been in arrears.",
+        "I have never had any arrears, and never will!"
+      ]
+    },
+    {
+      "intent": "future plans",
+      "label": "higher studies",
+      "examples": [
+        "Future plans",
+        "What do you plan to do in the future?",
+        "Got any plans for later?",
+        "Higher studies",
+        "Are you planning for higher studies?",
+        "After college",
+        "What will you do after college?",
+        "What will you do after getting a degree?"
+      ],
+      "response": "I have no plans for higher studies. I wish for a placement in a good company, but I have not decided my domain yet.",
+      "children": [
+        {
+          "intent": "higher studies",
+          "label": "higher studies",
+          "examples": [
+            "Are you planning for higher studies?",
+            "What are your plans for higher studies?",
+            "Do you want to pursue higher studies?"
+          ],
+          "response": "I have no plans for higher studies."
+        },
+        {
+          "intent": "placement",
+          "label": "placement",
+          "examples": [
+            "What will you do after college?",
+            "What will you do after getting a degree?",
+            "Are you looking for a job?",
+            "Are you looking for placement?"
+          ],
+          "response": "I wish for a placement in a good company, but I have not decided my domain yet."
+        }
+      ]
+    },
+    {
+      "intent": "field of interest",
+      "label": "machine learning, deep learning",
+      "examples": [
+        "What are you interested in?",
+        "Technical interests?",
+        "What is your field of interest?",
+        "Is something your field of interest?"
+      ],
+      "response": "I am particularly interested in machine learning and deep learning, but I do not dislike other domains."
+    },
+    {
+      "intent": "hobbies",
+      "label": "blender, badminton",
+      "examples": [
+        "What do you in your free time?",
+        "Hobbies",
+        "What are your hobbies?",
+        "Recreational activities",
+        "Blender",
+        "Do you use blender?",
+        "What do you do in blender?",
+        "How long have you been using blender?",
+        "Renders",
+        "Can I see your renders?",
+        "Show me your renders",
+        "Badminton",
+        "How long have you been playing badminton?",
+        "How good are you at badminton?",
+        "Drawing",
+        "What do you draw?",
+        "Since when have you been drawing?",
+        "Can I see your drawings?",
+        "Show me your drawings"
+      ],
+      "response": "I am a 3D enthusiast primarily working on the free open source software Blender. I also play badminton, and draw sometimes.",
+      "children": [
+        {
+          "intent": "blender",
+          "label": "blender",
+          "examples": [
+            "Blender",
+            "Do you use blender?",
+            "What do you do in blender?",
+            "How long have you been using blender?",
+            "How experienced are you in blender?",
+            "What have you made with blender?",
+            "How do you use blender?",
+            "Can you show me your renders?",
+            "Any good renders?",
+            "Renders",
+            "Can I see your renders?",
+            "Show me your renders"
+          ],
+          "response": "I have been using blender for since 2021. I am primarily a hard surface modelling artist with some experience in environmental and interior design.",
+          "children": [
+            {
+              "intent": "blender time",
+              "label": "blender",
+              "examples": [
+                "When did you start using blender?",
+                "How long have you been using blender?",
+                "Since when have you been using blender?"
+              ],
+              "response": "I have been using blender since 2021."
+            },
+            {
+              "intent": "blender render",
+              "label": "blender",
+              "examples": [
+                "What have you made with blender?",
+                "How do you use blender?",
+                "Can you show me your renders?",
+                "Any good renders?",
+                "Renders",
+                "Can I see your renders?",
+                "Show me your renders"
+              ],
+              "response": "Here is some of my work. <br> <img src = '/static/mountain_tent.jpg' onclick = 'requestFullscreen(this)'> <br> <img src = '/static/forest_tent.png' onclick = 'requestFullscreen(this)'> <br> <img src = '/static/my_room.png' onclick = 'requestFullscreen(this)'>"
+            },
+            {
+              "intent": "blender experience",
+              "label": "experience",
+              "examples": [
+                "How experienced are you in blender?",
+                "How good are you at blender?",
+                "How good are you at 3D modelling?"
+              ],
+              "response": "I have been using blender for over 4 years, so I consider myself to be decent."
+            }
+          ]
+        },
+        {
+          "intent": "badminton",
+          "label": "badminton",
+          "examples": [
+            "Badminton",
+            "How long have you been playing badminton?",
+            "Since when have you been playing badminton?",
+            "How experienced are you at badminton?"
+          ],
+          "response": "I have been playing badminton since I was 12, but I have had very long breaks in between. Now it is a part of my daily life, and I consider myself an above average player.",
+          "children": [
+            {
+              "intent": "badminton time",
+              "label": "badminton",
+              "examples": [
+                "How long have you been playing badminton?",
+                "Since when have you been playing badminton?"
+              ],
+              "response": "I have been playing badminton since I was 12, but I have had very long breaks in between. Now it is a part of my daily life."
+            },
+            {
+              "intent": "badminton experience",
+              "label": "experience",
+              "examples": [
+                "How good are you at badminton?",
+                "How experienced are you at badminton?"
+              ],
+              "response": "I consider myself an above average player."
+            }
+          ]
+        },
+        {
+          "intent": "drawing",
+          "label": "drawing",
+          "examples": [
+            "Drawing",
+            "What do you draw?",
+            "Since when have you been drawing?",
+            "Can I see your drawings?",
+            "Show me your drawings"
+          ],
+          "response": "I draw rarely. It used to be my favourite hobby in the past, until I discovered Blender.",
+          "children": [
+            {
+              "intent": "drawing time",
+              "label": "drawing",
+              "examples": [
+                "How long have you been drawing?",
+                "Since when have you been drawing?"
+              ],
+              "response": "I have been drawing since I was 10 years old."
+            },
+            {
+              "intent": "drawing experience",
+              "label": "drawing",
+              "examples": [
+                "How good are you at drawing?",
+                "How experienced are you at drawing?"
+              ],
+              "response": "I am really not sure... it turns out really well sometimes and really bad some other times."
+            },
+            {
+              "intent": "see drawings",
+              "label": "see drawings",
+              "examples": [
+                "Can I see your drawings?",
+                "Show me your drawings",
+                "What have you drawn?"
+              ],
+              "response": "Here are some of my drawings. <br> <img src = '/static/holter.jpg' onclick = 'requestFullscreen(this)'> <br> <img src = '/static/stark.jpg' onclick = 'requestFullscreen(this)'> <br> <img src = '/static/kratos.jpg' onclick = 'requestFullscreen(this)'>"
+            }
+          ]
+        }
+      ]
+    },
+    {
+      "intent": "dream",
+      "label": "dream",
+      "examples": [
+        "What is your dream?",
+        "Dream job?",
+        "What do you want to be?",
+        "What is your ambition?",
+        "What do you aspire to be?"
+      ],
+      "response": "I would like to be placed among my favourite companies, which are Google, Nvidia and Tesla, but I have no specific dream job.",
+      "children": [
+        {
+          "intent": "dream company",
+          "label": "google, nvidia, tesla",
+          "examples": [
+            "What is your dream company?",
+            "Dream company?",
+            "Which company do you want to work for?",
+            "What is your favourite company?",
+            "What companies do you like?",
+            "Where do you want to work?",
+            "Where do you want to be placed?",
+            "What is your dream job?"
+          ],
+          "response": "My favourite companies are Google, Nvidia and Tesla. I would like to be placed among them."
+        },
+        {
+          "intent": "dream",
+          "label": "artificial intelligence",
+          "examples": [
+            "What is your dream?",
+            "What do you want to achieve?",
+            "What is something you want to do?"
+          ],
+          "response": "My dream is to revolutionize the field of artificial intelligence and create industry standard applications."
+        }
+      ]
+    },
+    {
+      "intent": "work experience",
+      "label": "work",
+      "examples": [
+        "Do you have work experience?",
+        "Any jobs before?",
+        "Have you worked before?",
+        "Your job history?",
+        "Have you worked in a job before?",
+        "Experience",
+        "What is your experience?",
+        "About your experience"
+      ],
+      "response": "I haven't done any official jobs yet. Just learning and building stuff for now."
+    },
+    {
+      "intent": "internships",
+      "label": "internships",
+      "examples": [
+        "Have you done any internships?",
+        "Internships?",
+        "Have you interned anywhere?",
+        "Any internships?"
+      ],
+      "response": "I haven't done any internships yet. I'm learning stuff for now."
+    },
+    {
+      "intent": "competitions",
+      "label": "competitions",
+      "examples": [
+        "Have you participated in any competitions?",
+        "Competitions?",
+        "Have you competed in anything?",
+        "Any competitions?",
+        "Hackathons?",
+        "Have you participated in any hackathons?",
+        "Have you competed in any hackathons?",
+        "Any hackathons?",
+        "Have you qualified for a hackathon?"
+      ],
+      "response": "I have participated in a few hackathons, but I haven't won any yet. I'm still learning and improving."
+    },
+    {
+      "intent": "skills",
+      "label": "skills",
+      "examples": [
+        "What are your skills?",
+        "Skills?",
+        "What are you good at?",
+        "What are your strengths?",
+        "What are your abilities?",
+        "What can you do?"
+      ],
+      "response": "I am skilled in the languages Python, Rust, C++, Java, HTML, CSS, JavaScript, Java, Kotlin, Golang and Haskell. I am familiar with the tech stack Vulkan, OpenGL, Pytorch, Tensorflow, OpenCV, Django, FastAPI, Tokio, Rayon, WebGPU and CUDA."
+    },
+    {
+      "intent": "specialization",
+      "label": "specialization",
+      "examples": [
+        "What is your specialization?",
+        "Specialization?",
+        "What are you specialized in?",
+        "What is your major?"
+      ],
+      "response": "I am specialized in Computer Science, with a focus on AI/ML and GPU rendering. I am also interested in low-level graphics programming."
+    },
+    {
+      "intent": "certifications",
+      "label": "certifications",
+      "examples": [
+        "Have you done any certifications?",
+        "Certifications?",
+        "Have you completed any courses?",
+        "Any certifications?",
+        "Have you done any online courses?"
+      ],
+      "response": "I have completed a few online courses, but I haven't received any certifications yet."
+    },
+    {
+      "intent": "languages",
+      "label": "languages",
+      "examples": [
+        "Languages?",
+        "What languages do you know?",
+        "How many languages do you know?",
+        "Can you speak language?",
+        "What languages can you speak?",
+        "What languages are you fluent in?"
+      ],
+      "response": "I am fluent in English, Tamil and Telugu, and somewhat fluent in Hindi and Malayalam."
+    },
+    {
+      "intent": "merits and demerits",
+      "label": "merits and demerits",
+      "examples": [
+        "What are your merits?",
+        "What are your demerits?",
+        "What are your merits and demerits?",
+        "What are your strengths and weaknesses?",
+        "What are your pros and cons?",
+        "What are your good and bad points?",
+        "What are your good and bad qualities?"
+      ],
+      "response": "I am a very curious person, and I love to learn new things. I am also very stubborn, and I don't give up easily. My demerit is that I am not very good at communicating with people, and I tend to be a bit shy.",
+      "children": [
+        {
+          "intent": "merits",
+          "label": "merits",
+          "examples": [
+            "What are your merits?",
+            "What are your strengths?",
+            "What are your good points?",
+            "What are your good qualities?"
+          ],
+          "response": "I am a very curious person, and I love to learn new things. I am also very stubborn, and I don't give up easily."
+        },
+        {
+          "intent": "demerits",
+          "label": "demerits",
+          "examples": [
+            "What are your demerits?",
+            "What are your weaknesses?",
+            "What are your bad points?",
+            "What are your bad qualities?"
+          ],
+          "response": "My demerit is that I am not very good at communicating with people, and I tend to be a bit shy."
+        }
+      ]
+    },
+    {
+      "intent": "research papers",
+      "label": "research papers",
+      "examples": [
+        "Have you published any research papers?",
+        "Research papers?",
+        "Have you written any research papers?",
+        "Any research papers?",
+        "Have you written any papers?",
+        "Are you researching on something?",
+        "Are you doing any research?",
+        "Are you working on any research?"
+      ],
+      "response": "I have not published any research papers yet, but I am working on a few."
+    },
+    {
+      "intent": "projects",
+      "label": "projects",
+      "examples": [
+        "What projects are you working on?",
+        "What are your projects?",
+        "What are you working on?",
+        "What is your current project?",
+        "What is your latest project?",
+        "What are your current projects?",
+        "What are your latest projects?",
+        "How do you even work?",
+        "How do you function?",
+        "How are you able to chat?",
+        "What is your technology?",
+        "What is your tech stack?",
+        "What is your architecture?",
+        "How are you built?",
+        "How are you made?",
+        "What is your framework?",
+        "About your tech stack",
+        "About your technology",
+        "About your working",
+        "About Amber AI",
+        "About Amber",
+        "About you",
+        "How does Amber work?",
+        "How does Amber function?",
+        "Arx-Net",
+        "What is Arx-Net?",
+        "How does Arx-Net work?",
+        "What is the tech stack of Arx-Net?",
+        "How is arx-net built?",
+        "What is the architecture of Arx-Net?"
+      ],
+      "response": "I am currently working on Amber AI. My other projects include Arx-Net, an open source graph visualization tool, <project 2>, <project 3>, and <project 4>.",
+      "children": [
+        {
+          "intent": "Amber AI",
+          "label": "Amber AI",
+          "examples": [
+            "How does Amber AI even work?",
+            "How does Amber AIfunction?",
+            "How is Amber AI able to chat?",
+            "What is Amber AI technology?",
+            "What is Amber AI tech stack?",
+            "What is Amber AI architecture?",
+            "How is Amber AI built?",
+            "How is Amber AI made?",
+            "What is Amber AI framework?",
+            "About Amber AI tech stack",
+            "About Amber AItechnology",
+            "About Amber AI working",
+            "About Amber AI",
+            "About Amber",
+            "How does Amber work?",
+            "How does Amber function?"
+          ],
+          "response": "I (Amber AI) am a chatbot created by Dwarakesh. For details on how I work, check out my source code and documentation on Github using <a href = 'https://www.github.com/Dwarakesh-V'> github.com/Dwarakesh-V </a>."
+        },
+        {
+          "intent": "Arx-Net",
+          "label": "Arx-Net",
+          "examples": [
+            "Arx-Net",
+            "What is Arx-Net?",
+            "How does Arx-Net work?",
+            "What is the tech stack of Arx-Net?",
+            "How is arx-net built?",
+            "What is the architecture of Arx-Net?"
+          ],
+          "response": "Arx-Net is an open source graph visualization tool. For details on how it works, check out the source code and documentation on Github using <a href = 'https://www.github.com/Dwarakesh-V'> github.com/Dwarakesh-V </a>."
+        }
+      ]
+    },
+    {
+      "intent": "contact",
+      "label": "contact",
+      "examples": [
+        "How can I contact you?",
+        "How can I reach you?",
+        "How can I get in touch with you?",
+        "How can I connect with you?",
+        "How to contact you?",
+        "Contact",
+        "Contact",
+        "Can I colab with you?",
+        "Can I collaborate with you?",
+        "Can I work with you?",
+        "Contact information",
+        "Gmail",
+        "Email",
+        "Github",
+        "LinkedIn",
+        "Social media",
+        "Social links",
+        "Social media accounts",
+        "What is your email?",
+        "What is your github?",
+        "What is your linkedin?",
+        "What is your social media?"
+      ],
+      "response": "You can contact me through Github on <a href = 'https://www.github.com/Dwarakesh-V'> github.com/Dwarakesh-V </a>, Email on <email link>, or LinkedIn on <a href = 'https://www.linkedin.com/in/vdwarakesh/'> linkedin.com/in/vdwarakesh </a>.",
+      "children": [
+        {
+          "intent": "email",
+          "label": "email",
+          "examples": [
+            "What is your email?",
+            "What is your email address?",
+            "Email",
+            "Gmail"
+          ],
+          "response": "My email is <email link>."
+        },
+        {
+          "intent": "github",
+          "label": "github",
+          "examples": [
+            "What is your github?",
+            "Github",
+            "Github link"
+          ],
+          "response": "My github is <a href = 'https://www.github.com/Dwarakesh-V'> github.com/Dwarakesh-V </a>."
+        },
+        {
+          "intent": "linkedin",
+          "label": "linkedin",
+          "examples": [
+            "What is your linkedin?",
+            "LinkedIn",
+            "LinkedIn link"
+          ],
+          "response": "My linkedin is <a href = 'https://www.linkedin.com/in/vdwarakesh/'> linkedin.com/in/vdwarakesh </a>."
+        }
+      ]
+    },
+    {
+      "intent": "favorite color",
+      "label": "favorite color",
+      "examples": [
+        "What is your favorite color?",
+        "Favorite color?",
+        "What color do you like?",
+        "Do you have a favorite color?",
+        "What colors do you like?"
+      ],
+      "response": "My favourite color is green, tied with black."
+    },
+    {
+      "intent": "favorite food",
+      "label": "favorite food",
+      "examples": [
+        "What is your favorite food?",
+        "Favorite food?",
+        "What food do you like?",
+        "Do you have a favorite food?",
+        "What foods do you like?"
+      ],
+      "response": "I like South Indian food in general, but I'm not very picky."
+    },
+    {
+      "intent": "favorite movie",
+      "label": "favorite movie",
+      "examples": [
+        "What is your favorite movie?",
+        "Favorite movie?",
+        "What movie do you like?",
+        "Do you have a favorite movie?",
+        "What movies do you like?"
+      ],
+      "response": "I dont have a single favourite movie. I like many movies, particularly the ones with a good story, like Harry Potter and Dangal."
+    },
+    {
+      "intent": "parents",
+      "label": "parents",
+      "examples": [
+        "What are your parents doing?",
+        "What is you father doing?",
+        "What is your mother doing?",
+        "What is your father's job?",
+        "What is your mother's job?",
+        "What do your parents do?",
+        "What do your parents work as?",
+        "What do your parents do for a living?"
+      ],
+      "response": "My father is a buisnessman, and my mother is a government employee."
+    },
+    {
+      "intent": "siblings",
+      "label": "siblings",
+      "examples": [
+        "Do you have any siblings?",
+        "Are you an only child?",
+        "Do you have a brother or sister?",
+        "How many siblings do you have?",
+        "What are your siblings doing?"
+      ],
+      "response": "I have no siblings. I am an only child."
+    },
+    {
+      "intent": "maritial status",
+      "label": "maritial status",
+      "examples": [
+        "Are you married?",
+        "Are you single?",
+        "Do you have a partner?",
+        "Are you in a relationship?",
+        "Do you have a girlfriend?",
+        "Are you dating?"
+      ],
+      "response": [
+        "I am as single as anyone could be.",
+        "Single at the start of the year. Single at the end of the year. Consistency is key."
+      ]
+    },
+    {
+      "intent": "interests",
+      "label": "interests",
+      "examples": [
+        "Is there someone you like?",
+        "Are you interested in anyone?",
+        "Do you have a crush?",
+        "Whom do you like?",
+        "Who is your crush?"
+      ],
+      "response": "There could be someone I like. Maybe I'm lying. Good luck finding out.",
+      "children": [
+        {
+          "intent": "who do you like",
+          "label": "who do you like",
+          "examples": [
+            "Who is it?",
+            "Whom do you like?",
+            "Who is your crush?"
+          ],
+          "response": "No comments."
+        }
+      ]
+    },
+    {
+      "intent": "age",
+      "label": "age",
+      "examples": [
+        "How old are you?",
+        "What is your age?",
+        "When were you born?",
+        "When is your birthday?",
+        "What year were you born?"
+      ],
+      "response": "I was born on November 2, 2005. Which means I am 19 years old.",
+      "children": [
+        {
+          "intent": "birth date",
+          "label": "birth date",
+          "examples": [
+            "When is your birthday?",
+            "What is your birth date?",
+            "When were you born?",
+            "What year were you born?"
+          ],
+          "response": "I was born on November 2, 2005."
+        },
+        {
+          "intent": "current age",
+          "label": "current age",
+          "examples": [
+            "How old are you?",
+            "What is your age?",
+            "How many years old are you?"
+          ],
+          "response": "I am 19 years old."
+        }
+      ]
+    },
+    {
+      "intent": "resume",
+      "label": "resume",
+      "examples": [
+        "Resume",
+        "Can I see your resume?",
+        "Show me your resume",
+        "Send me your resume"
+      ],
+      "response": "Here is my resume.<br><img src = '/static/resume_sample.webp' onclick = 'requestFullscreen(this)'>"
+    }
+  ]
+}

tree_data/portfolio_tree_data.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23dd6a8ccbe038d78153e6ce7f933979f9a48be617a835a31e29e738d8781cbe
+size 679822

util.py

@@ -0,0 +1,6 @@
+from random import randint
+def select_random_from_list(l):
+    if isinstance(l, list): 
+        return l[randint(0,len(l)-1)]
+    else:
+        return l

voice.py

@@ -0,0 +1,53 @@
+import os
+import wave
+import json
+import subprocess
+import tempfile
+from vosk import Model, KaldiRecognizer
+
+os.environ['KALDI_LOG_VERBOSE'] = '0'
+
+def transcribe_audio(audio_path):
+    """
+    Transcribes a WAV file using vosk-model-small-en-us-0.15. The file is converted to 16kHz mono PCM if necessary.
+    Parameters:
+        1. audio_path (str): Path to the input audio file.
+    """
+    model_path = "vosk-model-small-en-us-0.15"
+    if not os.path.exists(model_path):
+        raise FileNotFoundError(f"Vosk model not found at: {model_path}")
+
+    # Create temporary file for processed audio
+    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_wav:
+        tmp_path = tmp_wav.name
+
+    try:
+        # Convert input audio to required format using ffmpeg
+        subprocess.run([
+            "ffmpeg", "-y", "-i", audio_path,
+            "-ar", "16000", "-ac", "1", "-c:a", "pcm_s16le",
+            tmp_path
+        ], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, check=True)
+
+        # Open processed audio
+        wf = wave.open(tmp_path, "rb")
+        model = Model(model_path)
+        recognizer = KaldiRecognizer(model, wf.getframerate())
+
+        results = []
+        while True:
+            data = wf.readframes(4000)
+            if len(data) == 0:
+                break
+            if recognizer.AcceptWaveform(data):
+                res = json.loads(recognizer.Result())
+                results.append(res.get("text", ""))
+
+        final_res = json.loads(recognizer.FinalResult())
+        results.append(final_res.get("text", ""))
+
+        return " ".join(results).strip()
+
+    finally:
+        if os.path.exists(tmp_path):
+            os.remove(tmp_path)

vosk-model-small-en-us-0.15/README

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0cf286e4f7783306c5f6469b37db69228fb16803b03cae661edb2d7bba64ebb
+size 199

vosk-model-small-en-us-0.15/am/final.mdl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75370a0137f9daf8f469dedd7daa4513ae7a621f03240c6e512e2b50b656a7b6
+size 15962575

vosk-model-small-en-us-0.15/conf/mfcc.conf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e2228006d01d805ad1c267fee9f79709ca87ac51bd82b0e3f5c69ba543f0fc4
+size 131

vosk-model-small-en-us-0.15/conf/model.conf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f14cb1eeb07c762c371db648c6be688d347236155ca0f64fb13b6567a8ce81f
+size 290

vosk-model-small-en-us-0.15/graph/disambig_tid.int

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ad87cc166d0998f08f758f47a6223a120dfffcbee805c4849c3aa5e6bb3c0fc
+size 102

vosk-model-small-en-us-0.15/graph/phones/word_boundary.int

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da199d9c991e0e84681ddbb34627b915b26302d50a8fdaa23c51e2bc3a50b5c3
+size 1761

vosk-model-small-en-us-0.15/ivector/final.dubm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c5d7dd69d2122313baaf19f61f35dd3fa18b70c62ac0687e311e1c46e6daca7
+size 168048

vosk-model-small-en-us-0.15/ivector/final.ie

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f37faf90c375b9e4740b569398b5829ed9cc07d19be6d441f72c3b71d7efcc6
+size 8288887

vosk-model-small-en-us-0.15/ivector/final.mat

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddd83586dc5f928cda8738b922c85ffe38fc789cb5f9151a712ca12f37265382
+size 44975

vosk-model-small-en-us-0.15/ivector/global_cmvn.stats

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33be09afcc80059847a275c3d043b51f1ab954c7c2438ddbbf4745e8ba144ff9
+size 1080

vosk-model-small-en-us-0.15/ivector/online_cmvn.conf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2f3571754b64297cb7efb2e7ca3df61995c5a45fcbb97188f90613552bb2dfe
+size 95

vosk-model-small-en-us-0.15/ivector/splice.conf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f0c5f7c82d18eaf25d8bce470efa9f7741f88411fe428774bc0a9bb69a24756
+size 35