Juan Acosta - Resume

AI architect and data science leader who takes systems from prototype to production in regulated environments. Architected and built Project LOOM, a clinical trial documentation system validated and tested by Duke Clinical Research Institute, alongside production multi-agent systems for oncology summarization and payor appeals. Previously led a 10+ person data science organization and built a startup's professional services arm into a profitable line of business, delivering ML that served 2M+ daily predictions across 10M+ Medicare and Medicaid lives and contributed to $81M+ in Medicare Trust Fund savings.

Professional Experience

Founder & Principal Consultant

Acosta-Li LLC

October 2024 - Present

Lead development of Lucy, a multi-agent strategy advisor for postsecondary transformation (Google ADK, Azure OpenAI, A2A) - institution matching across 7,000+ NCES records, transformation-stage inference, and semantic retrieval over a curated research corpus. AI tools for education at the Bill & Melinda Gates Foundation, subcontracted through Alithi Management Consulting (December 2025 - present).
Advise Sylvan Labs on applied AI (January 2026 - present); engagement expanded into hands-on delivery in 2026.
Built and maintain Atlas Mineral for Alpax, a B2B SaaS platform for precious metal order reporting and tracking running $1M+ monthly.
Prior engagements include SpyCloud (supply chain attack vulnerability), Capital Pulse (financial risk and forecasting), and Beheld. Manage a bench of 4 data scientists and 2 project managers.

Director of Data Science

Health Universe

October 2024 - July 2026

Began as a client engagement through Acosta-Li; full-time April 2025 - July 2026.

AI architect and primary developer of Project LOOM, an i-Cubed Center clinical trial documentation system validated and tested by Duke Clinical Research Institute - generates protocols, manuals of procedures, safety plans, and recruitment materials as one consistent set, detecting contradictions introduced by edits.
Built an oncology summarization system that ingests complete patient records, disambiguates among multiple primaries, and assembles a full clinical history per cancer.
Built an agent-to-agent clinical appeals system that submits and appeals pathology invoices to payors, pre-screening justifications for likely denials and surfacing supporting evidence from patient history.
Led pre- and post-sales client engagements; fine-tuned LLMs for medical NER supporting patient summaries, prior authorizations, and trial recruitment.

Senior Manager, Data Operations

Babson Diagnostics

April 2024 - August 2024

Client of ClosedLoop.ai from 2022; joined full-time in 2024 to complete the program.

Developed and operationalized the algorithmic platelet assay - real-time ML (Docker, MongoDB, Python) taking small-volume platelet counting from research prototype to commercial launch - with integrations linking Sysmex analyzers, LIS, and Salesforce; led pipelines for order tracking, sample lifecycle, and financial monitoring at 1,000+ daily samples.

Manager, Data Science

ClosedLoop.ai

June 2021 - March 2024

Reported directly to the CEO as a non-voting member of the executive team; no CTO in place.

Led the data science organization (10+ data scientists, engineers, and project managers) and built the professional services team into a profitable line of business; technical advisor for 30+ enterprise clients from pre-sales through deployment.
Architected distributed ML pipelines on AWS (SageMaker, Jenkins CI/CD) powering 2M+ daily predictions; built EHR/FHIR to SQL pipelines with feature engineering, cohort logic, and drift and bias monitoring.
Designed predictive models for disease progression and admissions contributing to $81M+ in Medicare savings; platform took Best in KLAS for healthcare AI/ML, 2021-2023.

Software Engineer

Perfaware

July 2020 - June 2021

Built scalable backend systems for pharmacy curbside pickup and order management during COVID-19.

Data Researcher

Project SEED, University of Texas at Austin

2018 - 2020

Developed and deployed a regression model for pediatric prehypertension prediction integrated into clinical decisions.

Selected Recognition and Publications

Education

Technical Skills

GenAI & agentic systems: Multi-agent orchestration (A2A protocol, Google ADK, MCP, LangGraph), LLM routing (LiteLLM), RAG and hybrid retrieval, fine-tuning (medical NER), evals (inspect-ai), observability (Langfuse), inference cost optimization

Healthcare data: FHIR, HL7, X12 EDI 837/835, ICD-10, CPT/HCPCS, LOINC, SNOMED, RxNorm, EHR integration, HIPAA/PHI-aware design

ML & data science: PyTorch, Hugging Face Transformers, scikit-learn, TensorFlow, Spark, SHAP, drift and bias monitoring

Platforms & cloud: Python, TypeScript, SQL, R, Rust — FastAPI, Next.js, React, Svelte — PostgreSQL, MongoDB, Snowflake, AWS (SageMaker, Lambda, S3), Azure (Azure OpenAI, Cosmos DB), GCP (Vertex AI, BigQuery), Docker, Kubernetes, GitHub Actions

Also

Co-Founder & Director of Coffee

Mù Coffee ATX

February 2025 - Present

Co-founded a specialty coffee shop and importer of Chinese coffee, running sourcing and import logistics end to end; oversee operations, finance, and performance tracking.