🤖 assistant

My portfolio's AI backend. Completely useless to you. It only talks about me — and honestly, it does a better job of it than I do.

What is this

A RAG backend built with FastAPI that answers questions about Ignacio Figueroa.

Ask it about my projects → it answers. Ask it about the meaning of life → it tells you to contact me directly. Scope enforced at the prompt level, not by good faith.

Stack

Tool	Why
FastAPI	Because Flask is for people who haven't decided yet
Gemini 2.5 Flash	Replies faster than I do in a standup
LangChain	Tool calling and streaming without hand-rolling the loop
pgvector + Neon	Vector search without running a database cluster in my apartment
HuggingFace Inference API	Embeddings without melting the server
Payload CMS	Where the projects and experience actually live
uv	Dependency management from this century
Docker	"Works on my machine" — shipped

How it works

User asks something about Nacho
          ↓
Question converted to an embedding vector
          ↓
Cosine similarity search against Neon (pgvector)
          ↓
Most relevant context chunks injected into the prompt
          ↓
Gemini answers — calling tools (projects, experience,
contact, job-match) against Payload CMS when it needs
live data
          ↓
Response streams back token by token
          ↓
You know more about me than my own mother does

Endpoints

Method	Path	What it does	Rate limit
GET	/	ASCII banner and uptime. Peak engineering.	—
POST	/chat	Ask something about Nacho. Get a streamed answer.	10/min
GET	/portfolio/projects	Published projects, ?locale=en|es	—
GET	/portfolio/experience	Work experience, ?locale=en|es	—
POST	/portfolio/summarize	Summarize a Payload Lexical body	5/min
GET	/docs	Swagger UI — because we're professionals	—

POST /chat returns a raw text/plain token stream (not SSE frames), so read it with fetch + response.body.getReader().

Run locally

# Install dependencies
uv sync

# Build the knowledge base from Payload CMS (full rebuild, safe to re-run)
uv run python scripts/ingest.py

# Start the server
uv run uvicorn main:app --reload

The server runs at http://localhost:8000 by default. Swagger UI at http://localhost:8000/docs — use it.

Environment variables

DATABASE_URL=        # Neon connection string with pgvector enabled
GEMINI_API_KEY=      # or GOOGLE_API_KEY, which takes precedence
HF_TOKEN=            # HuggingFace Inference API token
PAYLOAD_CMS_URL=     # Payload CMS API base, e.g. https://site.com/api
FRONTEND_URL=        # Allowed CORS origins, comma-separated

License

MIT. Do whatever you want with it. Just don't ask the assistant for code — it'll redirect you to me, and I'm busy.