imran-decoder/pathway

1import os2import re3import json4import uvicorn5from typing import List, Optional, Dict, Any6from fastapi import FastAPI, HTTPException7from pydantic import BaseModel8from dotenv import load_dotenv9from pymongo import MongoClient10from bson import ObjectId, errors as bson_errors11from langchain_community.graphs import Neo4jGraph12from huggingface_hub import InferenceClient13 14# --- 1. Configuration & Setup ---15load_dotenv()16 17app = FastAPI(title="NSQF Career Pathway API")18 19# Verify Environment Variables20required_vars = ["NEO4J_URI", "NEO4J_USERNAME", "NEO4J_PASSWORD", "MONGO_URI", "HF_TOKEN"]21missing_vars = [var for var in required_vars if not os.getenv(var)]22if missing_vars:23    print(f"⚠️  CRITICAL WARNING: Missing environment variables: {missing_vars}")24    print("⚠️  Ensure .env file is present. Exiting...")25    exit(1)26 27# Initialize Clients28try:29    print("⏳ Connecting to Neo4j...")30    graph = Neo4jGraph(31        url=os.getenv("NEO4J_URI"),32        username=os.getenv("NEO4J_USERNAME"),33        password=os.getenv("NEO4J_PASSWORD")34    )35 36    print("⏳ Connecting to MongoDB...")37    client = MongoClient(os.getenv("MONGO_URI"))38    db = client["career-advisor"]39    users_collection = db["userdetails"]40 41    print("⏳ Connecting to Hugging Face...")42    hf_client = InferenceClient(43        model="meta-llama/Llama-3.1-8B-Instruct",44        token=os.getenv("HF_TOKEN")45    )46    print("✅ All systems initialized.")47 48except Exception as e:49    print(f"❌ Global Init Error: {e}")50    exit(1)51 52 53# --- 2. Data Models ---54class UserRequest(BaseModel):55    user_id: str56 57class PathwayStep(BaseModel):58    step_order: int59    nsqf_level: str60    role_title: str61    description: str62    skills_required: List[str]63    user_alignment: str64 65class CareerPathwayPlan(BaseModel):66    target_sector: str67    user_fit_summary: str68    steps: List[PathwayStep]69 70class APIResponse(BaseModel):71    status: str72    data: CareerPathwayPlan73 74 75# --- 3. Helper Functions ---76def extract_keywords(text: str) -> List[str]:77    """Extracts significant keywords from a string."""78    if not text: return []79    STOPWORDS = {80        "and", "or", "the", "in", "of", "for", "with", "a", "an", "to", "at",81        "by", "on", "is", "are", "was", "were", "be", "been", "being",82        "sector", "industry", "field", "area", "services", "management", "development"83    }84    clean_text = re.sub(r'[^a-zA-Z0-9\s]', '', text.lower())85    words = clean_text.split()86    keywords = [w for w in words if w not in STOPWORDS and len(w) > 2]87    return list(set(keywords))88 89NSQF_MAPPING = {90    "skills": {91        "Python": ["IT-ITeS", "Data Science"],92        "Java": ["IT-ITeS"],93        "Teaching": ["Education"],94        "Nursing": ["Healthcare"],95        "Welding": ["Manufacturing"]96    },97    "interestSectors": {98        "Technology": ["IT-ITeS"],99        "Health": ["Healthcare"],100        "Construction": ["Infrastructure"]101    },102    "qualifications": {103        "B.Tech": ["IT-ITeS", "Engineering"],104        "MBBS": ["Healthcare"],105        "B.Ed": ["Education"]106    }107}108 109def resolve_user_nsqf_sectors(user_doc: Dict) -> List[str]:110    """Maps user attributes to official NSQF Sectors."""111    matched_sectors = set()112    for skill in user_doc.get("skills", []):113        matched_sectors.update(NSQF_MAPPING["skills"].get(skill, []))114    for sec in user_doc.get("interestSectors", []):115        matched_sectors.update(NSQF_MAPPING["interestSectors"].get(sec, []))116    117    highest_q = user_doc.get("education", {}).get("highestQualification", "")118    if highest_q:119        matched_sectors.update(NSQF_MAPPING["qualifications"].get(highest_q, []))120        121    return list(matched_sectors)122 123 124# --- 4. API Endpoint ---125@app.post("/generate-pathway", response_model=APIResponse)126async def generate_pathway(payload: UserRequest):127    target_user_id = payload.user_id128 129    # --- STEP A: Fetch User from MongoDB ---130    if not ObjectId.is_valid(target_user_id):131        raise HTTPException(status_code=400, detail="Invalid MongoDB ObjectId format")132 133    try:134        user_doc = users_collection.find_one({"_id": ObjectId(target_user_id)})135    except bson_errors.InvalidId:136        raise HTTPException(status_code=400, detail="Invalid User ID")137 138    # Fallback for testing if ID doesn't exist139    if not user_doc:140        print(f"⚠️ User {target_user_id} not found. Using Mock Data.")141        user_doc = {142            "_id": target_user_id,143            "skills": ["Python", "Data Analysis", "SQL"],144            "interestSectors": ["Technology"],145            "education": {"highestQualification": "B.Tech"}146        }147 148    # --- STEP B: Determine Sector Context ---149    user_nsqf_sectors = resolve_user_nsqf_sectors(user_doc)150    151    # --- STEP C: Neo4j Query (Find Top Sector) ---152    query_sector = f"""153    MATCH (u:User {{mongoId: "{target_user_id}"}})-[r:RECOMMENDED_FOR]->(q:Qualification)-[b:BELONGS_TO_SECTOR]->(s:Sector)154    WITH s, q, r155    ORDER BY r.rank DESC156    RETURN s.name AS sector, head(collect(q.title)) AS top_title, max(r.rank) AS rank157    ORDER BY rank DESC158    LIMIT 1159    """160 161    try:162        result_sector = graph.query(query_sector)163    except Exception as e:164        print(f"Neo4j Error: {e}")165        result_sector = []166 167    # Fallback if Neo4j returns nothing168    if result_sector:169        top_sector = result_sector[0]['sector']170        top_title = result_sector[0]['top_title']171    else:172        top_sector = "IT-ITeS" # Default173        top_title = "Software Developer"174 175    # --- STEP D: Neo4j Query (Get Pathway Data) ---176    sector_keywords = extract_keywords(top_sector)177    title_keywords = extract_keywords(top_title)178    all_keywords = list(set(sector_keywords + title_keywords))179    180    sanitized_keywords = [kw.replace("'", "") for kw in all_keywords]181    182    # Build dynamic WHERE clause183    if sanitized_keywords:184        where_clause = " OR ".join([f"toLower(q.title) CONTAINS '{kw}'" for kw in sanitized_keywords])185    else:186        where_clause = "1=1"187 188    query_pathway = f"""189    MATCH (q:Qualification)-[:BELONGS_TO_SECTOR]->(s:Sector)190    WHERE s.name = "{top_sector}" AND ({where_clause})191    RETURN q.title AS title, q.nsqf_level AS nsqf_level, q.progression_pathway AS progression_pathway192    ORDER BY q.nsqf_level ASC193    LIMIT 10194    """195    196    try:197        pathway_data = graph.query(query_pathway)198    except Exception:199        pathway_data = []200 201    # --- STEP E: LLM Generation (Strict JSON) ---202    203    # 1. Define the Schema strictly204    json_schema = """205    {206      "career_pathway_plan": {207        "target_sector": "String",208        "user_fit_summary": "String",209        "steps": [210          {211            "step_order": 1,212            "nsqf_level": "String",213            "role_title": "String",214            "description": "String (Short description of role)",215            "skills_required": ["String", "String"],216            "user_alignment": "String (How well user fits)"217          }218        ]219      }220    }221    """222 223    # 2. Construct Prompt224    system_prompt = f"""225    You are a Career Counselor API Backend. 226    You DO NOT speak to the user directly. 227    You ONLY output valid JSON matching this schema exactly:228    {json_schema}229    230    Do not add Markdown formatting (like ```json). Just the raw JSON string.231    """232 233    user_prompt = f"""234    CONTEXT:235    User Skills: {user_doc.get('skills')}236    User Education: {user_doc.get('education', {}).get('highestQualification')}237    Target Sector: {top_sector}238    239    AVAILABLE QUALIFICATIONS (From Database):240    {pathway_data}241    242    TASK:243    Construct a logical career progression (Low NSQF -> High NSQF) using the qualifications above.244    Map the user's current skills to the appropriate starting point.245    """246 247    try:248        response = hf_client.chat.completions.create(249            model="meta-llama/Llama-3.1-8B-Instruct",250            messages=[251                {"role": "system", "content": system_prompt},252                {"role": "user", "content": user_prompt}253            ],254            max_tokens=1000,255            temperature=0.3, # Low temp for structure stability256            top_p=0.9257        )258 259        raw_content = response.choices[0].message.content.strip()260 261        # --- STEP F: JSON Cleanup & Validation ---262        263        # Remove Markdown wrappers if Llama adds them264        if "```json" in raw_content:265            raw_content = raw_content.split("```json")[1].split("```")[0].strip()266        elif "```" in raw_content:267            raw_content = raw_content.split("```")[1].split("```")[0].strip()268        269        # Parse270        parsed_json = json.loads(raw_content)271        272        # Validate structure wraps correctly273        if "career_pathway_plan" not in parsed_json:274            # Attempt to fix if LLM returned just the inner object275            parsed_json = {"career_pathway_plan": parsed_json}276 277        return {"status": "success", "data": parsed_json["career_pathway_plan"]}278 279    except json.JSONDecodeError:280        print(f"❌ JSON Parse Error. Raw Output:\n{raw_content}")281        raise HTTPException(status_code=500, detail="AI generated invalid JSON structure.")282    except Exception as e:283        print(f"❌ Processing Error: {str(e)}")284        raise HTTPException(status_code=500, detail=str(e))285 286 287if __name__ == "__main__":288    uvicorn.run(app, host="0.0.0.0", port=7860)