initial

app.py

@@ -1,64 +1,95 @@
+import os
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+import pandas as pd
+import torch
+import numpy as np
+from sentence_transformers import util
+import google.generativeai as genai
+import chromadb
+from langchain_chroma import Chroma
 
+# Load Gemini API key from environment variable
+genai.configure(api_key=os.environ["GEMINI_API_KEY"])
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+# Load and clean knowledge base
+df_kb = pd.read_json("XENO_Uganda_KnowledgeBase_Advisory.json")
+df_kb.dropna(subset=['Content'], inplace=True)
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+def prepare_documents(data):
+    documents, metadatas, ids = [], [], []
+    for item in data:
+        documents.append(f"Question: {item['Question']}\nAnswer: {item['Content']}")
+        metadatas.append({
+            "question": item["Question"],
+            "content": item["Content"],
+            "section": item.get("Section", ""),
+            "source": item.get("Source", ""),
+            "owner": item.get("Owner", ""),
+            "tag": item.get("Tag", "")
+        })
+        ids.append(item["ID"])
+    return documents, metadatas, ids
 
-    messages.append({"role": "user", "content": message})
+xeno_data_list = df_kb.to_dict('records')
+documents, metadatas, ids = prepare_documents(xeno_data_list)
 
-    response = ""
+# ChromaDB setup
+client = chromadb.PersistentClient(path="./xeno_db")
+collection_name = "xeno_collection"
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+try:
+    collection = client.get_collection(name=collection_name)
+except:
+    collection = client.create_collection(name=collection_name)
+    collection.add(documents=documents, metadatas=metadatas, ids=ids)
 
-        response += token
-        yield response
+vector_store = Chroma(client=client, collection_name=collection_name)
+retriever = vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 4})
 
+embedding_model = "models/embedding-001"
+llm_model_name = "models/gemma-3-4b-it"
 
+SYSTEM_PROMPT = """
+You are XENO Support Assistant, an AI-powered helpful and professional customer service representative.
+Use only the information provided in the knowledge base context to answer user queries.
+Do not hallucinate. If context doesn't contain relevant info, say so.
 """
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
 
+def generate_xeno_response(context, question):
+    model = genai.GenerativeModel(llm_model_name)
+    user_prompt = f"""{SYSTEM_PROMPT}
+    ### CONTEXT ###
+    {context}
+    ### QUESTION ###
+    {question}"""
+    response = model.generate_content(user_prompt)
+    return response.text.strip()
+
+def get_context_and_answer(query):
+    queried_results = retriever.invoke(query)
+    query_embedding = genai.embed_content(model=embedding_model,
+                                          content=query,
+                                          task_type="retrieval_query")['embedding']
+    cosine_scores = []
+    for doc in queried_results:
+        doc_embedding = genai.embed_content(model=embedding_model,
+                                            content=doc.page_content,
+                                            task_type="retrieval_document")['embedding']
+        cos_sim = util.cos_sim(torch.tensor(query_embedding).float(), torch.tensor(doc_embedding).float())[0][0].item()
+        cosine_scores.append(cos_sim)
+    context = ""
+    top_docs = sorted(zip(queried_results, cosine_scores), key=lambda x: x[1], reverse=True)[:2]
+    for i, (result, score) in enumerate(top_docs, 1):
+        context += f"Knowledge Entry {i}:\nQ: {result.metadata['question']}\nA: {result.metadata['content']}\n{'-'*40}\n"
+    return generate_xeno_response(context, query)
+
+iface = gr.ChatInterface(
+    fn=get_context_and_answer,
+    title="XENO Support Assistant",
+    description="Ask anything about XENO's financial services.",
+    theme="soft"
+)
 
 if __name__ == "__main__":
-    demo.launch()
+    iface.launch()
+