Institutional context
The Ifes project investigates a conversational assistant for searching, retrieving and analyzing intellectual-property assets. INPI is Brazil's National Institute of Industrial Property, a domain source and context rather than an employment relationship.
Investigated architecture
The solution combines Java and Spring Boot orchestration, a Python and FastAPI AI service, an Angular frontend, BM25 and embedding retrieval, provider adapters and answers grounded in retrieved context.
Preliminary evaluation
An initial eight-question set recorded 37.5% Exact Match with Groq and 50% with OpenAI. The sample is small, so these are early experimental results rather than definitive performance.