AI Engineering for Healthcare & Regulated Environments
We build production AI systems where accuracy alone is not enough — grounded in evidence, bounded by regulatory logic, and designed for human oversight and audit.
Active FDA contract · ECRI Institute · Veteran disability AI
Production AI in Regulated Environments
Systems built for environments where outputs carry regulatory weight — grounded evidence, human oversight, and full audit trails.
Medical Device Intelligence Platform
Five production pipelines over a 10,000+ device catalog — FDA 510(k)/PMA/De Novo and EU MDR/IVDR classification, horizon scanning from 72+ sources, attribute validation with confidence decomposition, EMBASE novelty mining, and an active FDA post-market AI surveillance contract.
Veteran Disability AI Systems — Trajector
Senior Lead Data Scientist role — production AI systems analyzing veteran claims files to surface evidence-based service connection findings for licensed VA advocates. Medical evidence retrieval with domain-specific embeddings, source-attributed nexus reasoning (99%+ traceability), full 38 CFR Parts 3 & 4 regulatory logic as deterministic guardrails, and a golden cohort regression harness across 253+ tests.
Most AI Systems Aren't Built for Where You're Deploying Them
Generic AI architecture works in environments where errors are recoverable. In healthcare and regulated workflows, they are not.
Outputs Without Sources
In regulated environments, 'the model said so' is not an answer. Every output needs evidence provenance — the specific documents, filings, or records that support the claim.
LLMs Applied to Rule-Bound Determinations
FDA classifications, coverage determinations, rating schedules — these are not ambiguous. Probabilistic inference where deterministic logic is required introduces failure modes that accuracy benchmarks cannot detect.
No Human Oversight Architecture
In regulated environments, AI is a force multiplier for expert review, not a replacement for it. Systems built without human-in-the-loop design fail at the exact edge cases regulators examine most closely.
Quality That Can't Be Measured
If you can't answer 'how do we know it's working?' with a structured evaluation harness and regression tests, you can't defend it to a regulator, a clinical leader, or an auditor.
Four Core Capabilities
Each designed for the specific architecture requirements of healthcare and regulated environments.
Evidence-Grounded AI Systems
RAG pipelines for regulated domains where outputs carry regulatory, clinical, or legal weight. Source authority weighting, per-output confidence decomposition, and provenance links built in from the start.
Learn more →Regulatory Logic Engineering
AI systems that encode regulatory frameworks as deterministic guardrails — LLMs handle the genuinely ambiguous tasks; statutes and regulations handle the rules.
Learn more →AI Evaluation & Quality Engineering
Structured evaluation harnesses, golden cohort regression testing, LLM-as-judge frameworks, and production observability for AI deployed in high-stakes workflows.
Learn more →Production AI Systems for Healthcare
Embedded AI engineering for healthcare and life sciences — full lifecycle ownership from R&D through production, with a weekly shipping cadence and measurement from day one.
Learn more →How We Build Regulated AI
Every layer answers a specific question: What is the evidence? What does the rule require? What is genuinely uncertain? How confident are we? Who reviewed it? How do we know it is still working?
Evidence Sources
FDA · PubMed · ClinicalTrials · Medical Records
Regulatory Logic
38 CFR · EU MDR Classes · FDA 510(k)/PMA
Evidence Retrieval
Provenance tracked · Authority-weightedSource-authority-weighted retrieval from FDA, PubMed, clinical registries, and domain corpora
Deterministic Guardrails
Regulatory logic · Confidence 1.0 where mandatedRegulatory rules applied as algorithmic logic — zero LLM for unambiguous determinations
Bounded LLM Reasoning
Output schema enforced · Confidence boundedLLMs handle genuinely ambiguous tasks only — output-constrained, evidence-anchored, schema-validated
Structured Output + Provenance
Per-output attribution · Audit trail storedEvery output carries source links, confidence scores, and the evidence chain behind the claim
Human Review Integration
Overrides stored · Expert judgment preservedAnalyst workflows, SME override paths, and feedback that routes back into the pipeline
Production Monitoring
Golden cohort gates deploys · Cost trackedGolden cohort regression gating, drift detection, and cost instrumentation from day one
Built by someone
who's done it.
Jorge Rivero is a PhD AI Engineer and enterprise machine learning leader with 10+ years building production AI in regulated environments — from FDA post-market surveillance and EU MDR device classification at ECRI, to 38 CFR regulatory logic for veteran disability systems at Trajector.
Named inventor on six US patents (assigned to Oracle Corporation), author of peer-reviewed research with 149+ citations, and an Erasmus Mundus scholar — Jorge brings academic rigor and hard-won production discipline to every engagement.
PredictIQ AI was founded to close the gap between AI ambition and AI execution in environments that can't afford to get it wrong.
View LinkedIn ProfileWorking on a High-Stakes AI Problem?
Tell us about the environment, the regulatory context, and what the failure mode looks like. We'll tell you honestly whether we can help.