Production
Government Institution (via Apiux Tech)
Production RAG system over 1M+ legal documents enabling structured drafting of legal petitions and tutela responses with full audit trails and human-review controls. Combined supervised fine-tuning with RAG over institutional records.
1M+Legal Documents
50→90%Drafting Accuracy
60%Review Time Saved
- vLLM
- RAG
- SFT
- On-Prem Deployment
- Python
- LangGraph
- PostgreSQL
ResearchCollaboration
University of Cincinnati — Research Collaboration
Multi-agent consensus architecture for hepatocellular carcinoma (HCC) clinical reasoning. Coordinates virtual specialist roles (hepatologist, oncologist, radiologist) with guideline-grounded retrieval. Led to ASCO 2026 abstract acceptance.
ASCO 2026Accepted Abstract
3Virtual Specialists
- LangGraph
- RAG
- Clinical Guidelines
- Multi-Agent
- Python
- LLM Evaluation
ProductionConfidential
FLAR — Fondo Latinoamericano de Reservas
Production document intelligence platform over 30K+ sensitive legal and institutional documents. Hybrid retrieval with cross-encoder reranking, automated SharePoint ingestion, full audit trails, and enterprise authentication.
30K+Documents Indexed
90%+QA Accuracy
1M+Legal Records
- pgvector
- Qdrant
- BGE-M3
- vLLM
- Azure
- Neo4j
- LangGraph
- Python
- FastAPI
Production
Apiux Tech — Internal Product
Ingests 200+ procurement opportunities per day from public sources. Generates structured qualification scores, business summaries, and ready-to-use proposal drafts in under 30 minutes — reducing commercial review latency from days to under an hour.
200+Opportunities/Day
<30 minTurnaround Time
Days→1hReview Latency
- Python
- LangGraph
- Claude API
- PostgreSQL
- Docker
- REST API
ProductionInternal
Apiux Tech
6+ specialized agent roles — product ownership, architecture, DevOps, QA, security, development — operating in a coordinated LangGraph graph. ~90% accuracy in human-reviewed code generation outputs during structured internal validation.
~90%Code Gen Accuracy
6+Specialized Agents
- LangGraph
- Claude API
- Python
- Docker
- Multi-Agent
- Structured Outputs
ResearchThesis
Universidad de los Andes — Undergraduate Thesis
Autonomous clinical interview agent for medical education. A LangGraph-based agentic loop simulating structured anamnesis interactions with medical students, evaluated on a 10-criterion rubric (clinical, conversational, safety dimensions) targeting inter-rater reliability of Cohen's κ ≥ 0.60.
10Evaluation Criteria
κ≥0.60IRR Target
- LangGraph
- Python
- OpenAI API
- Evaluation Framework
- Medical NLP