AI Agent · RAG
2026
Mockup / study
Haceb Agent — AgentSprint
Grounded (RAG) AI agent for the Haceb appliance lifecycle — AgentSprint / ReshapeX hackathon at EAFIT.
Why it was built
3.5-hour hackathon at EAFIT Medellín, teams of 3–4. The challenge: build a multi-agent anti-hallucination system for Haceb industrial catalogs, grounding every answer in a RAG vector base over the real product catalog.
The solution scored highest with the jury for accuracy and reliability on domain-specific queries.
How it works
- RAG vector base over the product catalog
- Validator sub-agent that cuts hallucinations
- Multi-provider: local Ollama, Groq and Gemini
- Streamlit UI
- Won on accuracy and reliability
Stack
Python
Streamlit
Gemini
RAG
Base vectorial