IGOR EDUARDO

Senior AI Product Engineer and Researcher

Researching production AI systems for healthcare.
880 tests. Zero hallucinations. 5,000 clinical queries evaluated.

Austin, TX

Stanford Design Thinking  ·  UC San Diego  ·  Y Combinator

Research

I study how AI systems can improve clinical decision-making in regulated healthcare environments.

My research spans hybrid retrieval architectures, citation verification, clinical safety systems, and multi-model inference optimization. Recent work involved engineering and evaluating two complete healthcare AI systems across 5,000 clinical queries with physician review panels.

Key findings: hybrid RAG (BM25 + semantic + Reciprocal Rank Fusion) achieving 91% retrieval precision. Deterministic citation verification eliminating hallucinations entirely. A 5-layer safety agent architecture with real-time drug interaction checking. Complexity-based model routing reducing inference computation by 73%.

100,000+ curated clinical documents indexed. 880 automated tests. Sub-2-second response latency. 10/10 audit score.

Evaluation Metrics

91%
Precision

Hybrid RAG. BM25 + semantic + RRF. Threshold calibrated at 0.60 across 5,000 clinical queries.

0%
Hallucinations

Deterministic citation verification. Every reference checked against retrieved documents. Under 5ms overhead.

5
Safety Layers

Emergency detection, PII scrubbing, drug interaction checks, evidence guardrails, regulatory disclaimers.

73%
Compute Reduction

Complexity-based model routing. 18,000 labeled queries. Automatic fallback chain.

880
Automated Tests

Vitest, Semgrep healthcare SAST (zero findings), Langfuse, OpenTelemetry. Full observability.

100k+
Documents

Curated clinical corpus. Authority-weighted: national protocols 2.0×, specialty 1.7×, international 1.3×.

Publications

arXiv preprint in preparation: Hybrid RAG for Clinical Decision Support

Open Source

Research components. MIT licensed.

llm-trust-layer

End-to-end LLM trust pipeline: routing + hybrid RAG + citation verification in a single call.

TypeScript · MIT 51 tests
hybrid-rag-pipeline

BM25 + semantic search + RRF fusion with authority boosting and recency weighting.

TypeScript · MIT 28 tests
citation-guard

Post-generation citation verification. Deterministic. Zero hallucinated references. <5ms overhead.

TypeScript · MIT 26 tests
model-router-chain

Complexity-based LLM routing with automatic fallback chain. 73% inference reduction.

TypeScript · MIT 18 tests
healthcare-ai-playbook

25 production-tested techniques for building AI in regulated healthcare environments.

MIT

Stack

Languages
TypeScript Python SQL
Frontend
React Next.js Tailwind
Backend
Node.js PostgreSQL pgvector Redis Supabase
AI / ML
RAG LLM routing Presidio RxNorm Embeddings Clinical NLP
Infrastructure
Railway Vercel AWS Docker GitHub Actions
Quality
Vitest Semgrep Langfuse OpenTelemetry Sentry
Compliance
HIPAA LGPD CFM 2.454/2026 GRADE FHIR R4

Experience

2024 — Present

Senior AI Product Engineer and Researcher

Austin, TX

Engineered and evaluated two healthcare AI systems. 10 integrated modules. Hybrid RAG over 100,000+ curated clinical documents. 5-layer safety agent. Deterministic citation verification (0% hallucination rate). Multi-model routing (73% inference reduction). 880 automated tests. Evaluated across 5,000 clinical queries. 10/10 due diligence audit.

2024 — Present

Ambassador

Gamma  ·  $2.1B valuation  ·  Remote

AI content creation platform.

2018 — 2024

Strategy, Operations and Technology

AVIV / OpenVC  ·  São Paulo

Portfolio growth and governance for R$1B+ AUM venture fund across 46 investees. Sourcing, due diligence, crisis PMO, board-level governance. IT infrastructure: CRM, data pipelines, analytics dashboards, DD automation. Key outcomes: Neoway exit ~$360M on B3. Grão Direto seed ($0.46M) to Series B ($15M).

2019 — 2020

Founder, CEO and Technical Lead  (Exited)

Kubrick.social

White-label campaign orchestration SaaS. Team of 6 engineers (Node.js, React, PostgreSQL, AWS). Enterprise clients: Uber, Rappi, Airbnb, Spotify, Diageo. 5-year license. Acquired by the largest PR conglomerate in Latin America.

Education

English (Fluent)  ·  Portuguese (Native)  ·  Spanish (Fluent)