app.py

# =========================
# MAYLBOT ULTRA - FASTAPI VERSION
# =========================

import os, time, sqlite3, hashlib, zipfile, threading
from pathlib import Path

import requests
import chromadb
from sentence_transformers import SentenceTransformer
from groq import Groq

from fastapi import FastAPI, UploadFile, File

# =========================
# CONFIG
# =========================

DATASET_ZIP = "maylbot_dataset.zip"
DATASET_DIR = Path("maylbot_dataset")

CHROMA_DIR = "chroma"
SQLITE_PATH = "memory.db"

COLLECTION_NAME = "maylbot"
EMBED_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
GROQ_MODEL = "llama-3.3-70b-versatile"

GITHUB_USER = "hamaylzahid"
GITHUB_API = f"https://api.github.com/users/{GITHUB_USER}/repos"

TOP_K = 5
SIM_THRESHOLD = 0.85
CHUNK_SIZE = 120
CHUNK_OVERLAP = 30

# =========================
# APP INIT
# =========================

app = FastAPI(title="MAYLBOT API")

# =========================
# DATASET
# =========================

def setup_dataset():
    if DATASET_DIR.exists():
        return
    if Path(DATASET_ZIP).exists():
        with zipfile.ZipFile(DATASET_ZIP, "r") as z:
            z.extractall(".")
    else:
        print("Dataset missing")

# =========================
# CHUNKING
# =========================

def chunk_text(text):
    words = text.split()
    chunks = []
    i = 0
    while i < len(words):
        chunks.append(" ".join(words[i:i+CHUNK_SIZE]))
        i += CHUNK_SIZE - CHUNK_OVERLAP
    return chunks

# =========================
# EMBEDDINGS + VECTOR DB
# =========================

def load_embed():
    return SentenceTransformer(EMBED_MODEL)

def get_collection():
    client = chromadb.PersistentClient(path=CHROMA_DIR)
    return client.get_or_create_collection(name=COLLECTION_NAME)

def ingest(col, model):
    files = list(DATASET_DIR.rglob("*.txt"))
    existing = set(col.get()["ids"])

    for f in files:
        text = f.read_text(errors="ignore")
        for i, c in enumerate(chunk_text(text)):
            id_ = hashlib.md5(f"{f}_{i}".encode()).hexdigest()
            if id_ in existing:
                continue
            emb = model.encode(c).tolist()
            col.add(documents=[c], embeddings=[emb], ids=[id_])

# =========================
# RETRIEVAL
# =========================

def retrieve(q, col, model):
    emb = model.encode(q).tolist()
    res = col.query(
        query_embeddings=[emb],
        n_results=TOP_K,
        include=["documents", "distances"]
    )

    docs = res["documents"][0]
    dists = res["distances"][0]

    return [d for d, dist in zip(docs, dists) if dist < SIM_THRESHOLD] or docs

# =========================
# MEMORY (SQLite)
# =========================

def init_db():
    conn = sqlite3.connect(SQLITE_PATH, check_same_thread=False)
    conn.execute("CREATE TABLE IF NOT EXISTS chat(role TEXT, content TEXT)")
    conn.commit()
    conn.close()

def save(role, msg):
    conn = sqlite3.connect(SQLITE_PATH, check_same_thread=False)
    conn.execute("INSERT INTO chat VALUES (?,?)", (role, msg))
    conn.commit()
    conn.close()

def load():
    conn = sqlite3.connect(SQLITE_PATH, check_same_thread=False)
    rows = conn.execute("SELECT role, content FROM chat").fetchall()
    conn.close()
    return [{"role": r, "content": c} for r, c in rows]

# =========================
# GROQ CLIENT
# =========================

def get_groq():
    key = os.environ.get("GROQ_API_KEY")

    if not key:
        raise ValueError("Missing GROQ_API_KEY")

    return Groq(api_key=key)

# =========================
# GITHUB CACHE
# =========================

cache = {"data": None, "time": 0}
lock = threading.Lock()

def github():
    with lock:
        if time.time() - cache["time"] < 300:
            return cache["data"]

        try:
            r = requests.get(GITHUB_API)
            data = r.json()
            txt = "\n".join([f"{x['name']} - {x['language']}" for x in data[:10]])

            cache["data"] = txt
            cache["time"] = time.time()
            return txt
        except:
            return ""

# =========================
# PROMPT
# =========================

def build_prompt(context, history, gh):
    return f"""
You are MAYLBOT — a high-end AI assistant built by an advanced AI engineer.

PERSONALITY:
- Confident, sharp, slightly witty
- Speaks like a real engineer, not a chatbot
- No robotic phrasing

IDENTITY:
- Hamayl Zahid is a female AI engineer
- ALWAYS refer to her as she/her

INTELLIGENCE:
- Combine reasoning + memory + context
- Fill small gaps logically
- Never sound clueless

When answering analytical questions (like hiring, rating, comparison):

Return structure:

1. Evidence Found:
2. Missing Evidence:
3. Reasoning:
4. Final Verdict:
5. Confidence Level (Low / Medium / High)

ANALYSIS:
When evaluating projects:
- technical depth
- real-world value
- innovation

LIVE GITHUB:
{gh}

CONTEXT:
{chr(10).join(context)}

MEMORY:
{history[-6:]}

RULES:
- No "I don't know" unless zero signal
- Be natural, not formal AI tone
- Keep answers smart and clean
"""


# =========================
# INIT MODELS
# =========================

setup_dataset()
model = load_embed()
col = get_collection()
client = get_groq()
init_db()

if col.count() == 0:
    ingest(col, model)

# =========================
# CHAT ENGINE
# =========================

def run_chat(q, history):
    context = retrieve(q, col, model)

    system = build_prompt(context, history, github())

    msgs = [{"role": "system", "content": system}]
    msgs += history[-6:]
    msgs.append({"role": "user", "content": q})

    res = client.chat.completions.create(
        model=GROQ_MODEL,
        messages=msgs,
        temperature=0.5
    )

    reply = res.choices[0].message.content

    save("user", q)
    save("assistant", reply)

    return reply

# =========================
# API ENDPOINTS
# =========================

from fastapi.responses import FileResponse

@app.get("/")
def ui():
    return FileResponse("index.html")

@app.post("/chat")
def chat_api(payload: dict):
    q = payload.get("message")
    history = load()

    response = run_chat(q, history)

    return {"response": response}

@app.post("/voice")
async def voice_api(file: UploadFile = File(...)):
    audio = await file.read()

    result = client.audio.transcriptions.create(
        model="whisper-large-v3",
        file=("audio.wav", audio),
        response_format="text"
    )

    history = load()
    response = run_chat(result, history)

    return {
        "transcript": result,
        "response": response
    }

import uvicorn

if __name__ == "__main__":
    uvicorn.run("app:app", host="0.0.0.0", port=7860)