README.md

metadata

title: ARIA — Arman's Portfolio Chatbot
emoji: 🛰️
colorFrom: gray
colorTo: blue
sdk: docker
app_port: 7860
pinned: false

ARIA — Arman's Resume Intelligence Assistant

Hand-coded RAG backend for Arman Adil Mangat's portfolio chatbot. No no-code tools.

Pipeline: knowledge markdown → ~300-token chunks → all-MiniLM-L6-v2 embeddings → FAISS (cosine) top-4 retrieval → Groq llama-3.1-8b-instant.

Stack: Flask · sentence-transformers · FAISS · Groq · Docker.

API

POST /chat
  body: { "message": str, "history": [{"role": "user"|"assistant", "content": str}] }
  resp: { "reply": str }

GET /health → {"status": "ok"}

CORS is restricted to the portfolio's GitHub Pages origin. Requests are rate-limited (~20/min per IP).

Secrets

One secret only: GROQ_API_KEY (Space → Settings → Variables and secrets). It never appears in code.

Note on cold starts

Free CPU Spaces sleep after ~48h of inactivity — the first request after a sleep can take ~20–30s while the container wakes. The portfolio frontend shows a "waking up the server" message for this case.