Igor Eduardo

Engineering AI
for complex systems.

I build systems powered by AI for environments where regulation, scale, reliability, and operational constraints shape every architectural decision. From research to production systems — where architecture matters more than frameworks.

Research · Production · Computer Vision · Retrieval · Edge AI · Clinical Systems
Everything published here is either publicly verifiable or intentionally generalized to respect client confidentiality and intellectual property.
Engineering Milestones

Engineering under real constraints, delivered under real deadlines.

Winner · webAI Build Challenge — Enterprise Track · 2026

SENTINEL

An on-device computer vision system, trained locally on Apple Silicon with MLX and YOLO26. Built for offline casualty triage — locally trained models, zero cloud dependency, designed for environments where connectivity cannot be assumed.

Winner · Antler ATX × Codex Hackathon · 2026

Eterna Bioscience

A research platform for peptide protocol tracking and real-world-evidence generation, built in 72 hours and designed around interoperable research data from day one.

Some engineering work is published in full. Some is intentionally published later — after the engineering reaches the standard I expect from public artifacts.

Principles
01

Constraints define the architecture, not the other way around.

02

Reproducibility beats intuition. Measure before optimizing.

03

Publish the reasoning, not only the code.

04

Some industries optimize for growth. I build for correctness.

Engineering

Where the hard part lives.
Retrieval · Portuguese clinical text

Hybrid RAG

BM25, dense retrieval, and Reciprocal Rank Fusion for Portuguese clinical text. Documented as an engineering lab — the decisions (ADRs), the experiments, and the architecture behind them, not just the code.

Problem
Clinical retrieval in Portuguese
Constraint
Ranking quality vs. serving cost
Decision
RRF over a reranker
Evidence
Paper · Dataset · Repository
Clinical AI · Architecture Case

DocMinds

A clinical AI platform for regulated healthcare, shown here as an architecture case — by choice. The engineering reasoning is public; the implementation that constitutes the product is not.

Problem
Clinical AI for physicians
Constraint
Compliance as a system requirement
Decision
Architecture-first design
Evidence
Architecture case

Public

  • Engineering principles
  • Design decisions
  • Architecture evolution
  • Compliance patterns

Private

  • Implementation
  • Production infrastructure
  • Retrieval strategies
  • Prompts & datasets

Research

How the work is validated.

Hybrid RAG for Portuguese Clinical Text

BM25 and dense retrieval are complementary, not competing, for Portuguese clinical retrieval. Evaluated across 500 clinical queries.

DOI
10.5281/zenodo.19686739

Brazilian-PHI

The first benchmark for detecting Brazilian PII in Portuguese clinical text — seven identifier types with mod-11 checksum validation.

DOI
10.5281/zenodo.20076543

Selected Engineering Challenges

Representative problems worked on across regulated and operational domains.

Clinical Imaging

  • Diagnostic-support systems built to operate under regulatory constraints

Clinical Privacy

  • De-identification that protects patient data while preserving its utility

Healthcare Operations

  • Reducing administrative friction across authorization and scheduling workflows

Investment Intelligence

  • Internal tooling for evaluating transactions and organizing due diligence

Operational Automation

  • CRM and lifecycle systems for customer operations

Media Processing

  • AI-assisted video pipelines for production environments

AI Security

  • Safeguards for clinical AI — PHI protection and prompt-injection defense around third-party inference

About

How it got here.

I build AI systems for problems where the engineering is the hard part. I train and evaluate models — including on-device inference on Apple Silicon — design retrieval and data pipelines, and publish the reasoning behind the systems whenever I can.

Before that, I founded and exited a technology company, then spent seven years in venture investing across healthcare, fintech, and SaaS. I also run an independent research capability (Cortexa) — computational pipelines, model training, and real-world-evidence work built from public data sources.

Roles Founder Research Engineer Advisory Acquired founder — Kubrick