AskUSP

A chatbot that answers questions about the University of São Paulo using freshly crawled, up-to-date university data — so answers reflect what's actually published today rather than what a model happened to memorise.

How it works

• Crawling — crawl4ai collects and cleans content from USP web pages.
• Storage — the crawled data is stored in PostgreSQL.
• Answering — an LLM (Anthropic Claude) answers user questions grounded in that data, with Instructor enforcing structured, validated outputs.
• API — served through FastAPI.

Stack

Python 3.13 · FastAPI · Anthropic Claude · Instructor · crawl4ai · PostgreSQL (psycopg) · Pydantic · loguru

Running

# Install dependencies (uses uv)
uv sync

# Configure environment (API keys, database URL) in config/
# then start the API
uv run -m app.main

Project layout

app/          FastAPI application and entry point
  models/     Data models and LLM response schemas
config/       Settings and environment configuration
database/     Schema and data-access code

Status

An experimental project exploring retrieval-grounded chat over live institutional data.