Andres Felipe Angel Imbacuan

LexAgent is a conversational legal assistant for Colombian law, focused on state contracting (Ley 80/1993, Ley 1150/2007), the Constitution, and SECOP public-procurement data. It solves a concrete problem: legal research in Colombia is slow, fragmented across official sites, and generic chatbots invent articles, laws, and citations when answering. LexAgent combines multi-turn chat with a hybrid RAG pipeline (Spanish full-text search + vector search + RRF fusion + deterministic rerank) over a curated corpus, so every factual claim carries a verifiable citation to the exact fragment (law > article > numeral). An agentic loop with function calling (buscar_normas, consultar_contratos) retrieves norms and real SECOP contracts before answering; guardrails enforce abstention when evidence is insufficient (0 obeyed injections and 5/5 effective abstentions on our 30-question eval), and responses stream over SSE. It also supports projects with per-project file uploads, conversation memory, cost tracking per answer, and auth with TOTP. Stack: NestJS + TypeScript API, React + Vite web, PostgreSQL + pgvector, Redis, Meta Model API (Muse Spark) for reasoning, and a dedicated embedding provider. Evaluated on 30 legal questions: 0 invented citations, 100% recall@8, 86.7% verified-or-partial support.