feat: add RAG chatbot to portfolio

[samridhsri]

Overview

Add an AI-powered chatbot to the portfolio site that answers questions about
my career, projects, and skills using RAG (Retrieval-Augmented Generation).

Stack: FastAPI · ChromaDB · sentence-transformers · Groq API (llama-3.1-8b-instant) · React

---

Phase 1 — Knowledge base ~1-2 hrs

• Create knowledge/ directory with 6 markdown files:
  about.md, experience.md, education.md, projects.md, skills.md, faqs.md
• Populate faqs.md with 10–15 recruiter-facing Q&As
• Keep each file under ~800 words for clean chunking

Deliverable: knowledge/ directory ready to ingest

---

Phase 2 — Ingest pipeline ~2-3 hrs

• Set up Python venv + requirements.txt
  (langchain, sentence-transformers, chromadb, python-dotenv)
• Write ingest.py: load markdowns → chunk (400 tok, 50 overlap) → embed → upsert ChromaDB
• Verify retrieval quality with manual similarity queries

Deliverable: chroma_db/ persisted locally, retrieval verified

---

Phase 3 — FastAPI backend ~3-4 hrs

• Scaffold main.py with POST /chat endpoint (accepts {message, history[]}, returns SSE)
• Build RAG chain: embed query → Chroma top-3 → build system prompt → stream via Groq
• Add CORS guard (portfolio domain only) + rate limiting (slowapi, 10 req/min per IP)
• Test locally with curl / Postman — verify streaming and context injection

Deliverable: backend running on localhost:8000, streaming verified

---

Phase 4 — React chat widget ~3-4 hrs

• Build ChatWidget.jsx: floating button → panel toggle, message list, input box
• Wire SSE streaming via fetch + ReadableStream (token-by-token append)
• Add 3 starter prompt chips ("What projects have you built?", etc.)
• Style to match existing portfolio theme (inherit CSS variables / Tailwind tokens)

Deliverable: widget live on local dev, end-to-end chat working

---

Phase 5 — Deploy ~1-2 hrs

• Add Dockerfile to backend repo (Railway auto-deploys from GitHub)
• Set GROQ_API_KEY and ALLOWED_ORIGIN env vars on Railway
• Update VITE_API_URL in frontend to Railway URL, redeploy portfolio
• Smoke test on production — streaming latency, CORS, 3–5 message flow

Deliverable: chatbot live on portfolio domain

---

Notes

• Commit chroma_db/ to repo (re-run ingest as Dockerfile startup step if it grows)
• Pass last 3 message pairs per request for session memory — no DB needed
• Re-run ingest.py whenever knowledge base files are updated