Senior Data Engineer | Healthcare AI | Trustworthy Data Systems

I design and build reliable data and AI systems for healthcare, research, and other high-stakes domains where trust matters.

Professionally, I’m a Senior Data Engineer with deep experience in data ingestion, analytics infrastructure, orchestration, and platform reliability. Independently, I’m building projects at the boundary of healthcare data engineering, applied AI, privacy-preserving systems, and human-centered tooling.

The common thread in my work is simple: complex systems should be understandable, reproducible, and useful to the people who depend on them.

I care about systems that:

• work reliably at scale,
• protect sensitive data,
• make uncertainty visible,
• and remain understandable to both technical and non-technical users.

Writing and notes: antonrasmussen.com

I’m currently building independent projects around:

• Verified / goal-directed agent harnesses for long-horizon healthcare data-engineering tasks — measuring and reducing false completion (claimed success that fails an independent oracle), with synthetic fixtures only
• Measurement choices that dominate apparent LLM reliability — prompt templates, scoring rules, calibration, and quantization effects in biomedical models
• Privacy-preserving data engineering and clinical / biomedical workflows
• Edge-ready and local-first AI systems
• Explainable, reproducible ML and experiment harnesses
• Agentic tooling for data platforms and research workflows

I’m especially interested in the practical middle ground between production engineering and applied research: turning messy data, fragile workflows, and vague questions into systems that can be tested, explained, and improved.

A long-term thread across this work is Careful Intelligence — my personal effort to explore trustworthy, privacy-aware AI systems that are useful in the real world, not just impressive in demos.

Languages Python, SQL, Bash, Java, TypeScript, JavaScript (comfortable learning new languages when the system demands it)

Data Engineering & Orchestration Dagster, Apache Spark, dbt, Pandas, Hadoop ecosystem

Cloud & Storage Google Cloud Platform, GCS, BigQuery, Datastore, Azure SQL, SQL Server, Teradata, Informix

Analytics, Visualization & Geospatial Tableau, Vega-Lite, Matplotlib, GeoPandas, OpenRefine

AI / ML Areas I Work Around LLM evaluation, model calibration, quantization, prompt stability, agent evaluation / verification gates, de-identification, NER, reproducible experiment harnesses

Tooling & Automation GitHub Actions scheduled pipelines · local-first SQLite tooling · CLI-first research utilities

Practices I Care About Idempotent pipelines · data validation · observability · schema evolution · reproducibility · de-identification · operational clarity · human review loops · independent verification over self-reported success

Cityblock Health | Feb 2024 – Present

• Design and evolve multi-source, multi-format ingestion pipelines for HIPAA-protected healthcare data
• Lead and support migration toward Dagster-based orchestration, improving reliability, extensibility, and developer velocity
• Build validation, profiling, quarantine, and review workflows that balance automation with human oversight
• Collaborate across analytics, clinical, implementation, and platform teams to make complex data systems more usable and trustworthy
• Work on operational visibility for ingestion workflows, including review paths for unknown, anomalous, or unconfigured inbound files

Cityblock Health | Feb 2022 – Feb 2024

• Built robust, idempotent ETL pipelines supporting analytics, operations, and clinical workflows
• Helped improve data quality, reliability, and maintainability across shared data infrastructure
• Supported ingestion and transformation patterns for healthcare data from multiple external sources

Walmart Global Tech | Dec 2019 – Feb 2022

• Developed large-scale analytics systems for pharmacy operations
• Worked with big-data platforms and production streaming/batch pipelines
• Built data-driven applications and dashboards supporting operational and analytical decision-making

Research prototype for measuring — and reducing — false completion in long-horizon agentic data-engineering work: the agent (or wrapper) reports success while an independent oracle shows the goal was not satisfied. Goals live outside the agent as machine-readable criteria; only an independent verifier can grant success. Framed around “when done isn’t.” Everything is synthetic fixtures (data, schemas, tickets, logs) — a measurement instrument, not a clinical product, and not a claim of production or HIPAA readiness.

CLI pipeline that turns a hand-authored shot manifest into a narrated video: Runway for text-to-video generation, ElevenLabs for narration, ffmpeg for local assembly — no manual editing, no UI work in either service. Cost-gated by design, with a free local rehearsal proving the assembly before any paid generation runs. Its own ~80-second teaser was produced end to end by the tool it demonstrates.

Preprint Quantization Effects on Biomedical LLM Reliability (Rasmussen & Qin) — AAAI 2026 Fall Symposium manuscript. Controlled evaluation of biomedical and general Mistral-7B variants on PubMed RCT under FP16, INT8, and INT4, showing that prompt templates and scoring rules can dominate apparent calibration and accuracy comparisons. Frozen course provenance (class-code protocol, not the paper source of truth): cs781-s26.

Python CLI for tracking what is actually new in UAP/UFO-related releases and coverage. Pulls from official and news sources, applies rule-based source trust logic, uses LLMs for summaries and novelty scoring, caches locally in SQLite, and runs a daily release watch with report email plus a weekly digest.

Automated pipeline over the official SAM.gov Opportunities API: configurable search profiles, rules-based explainable scoring, deduplication and amendment detection, document parsing, and daily email reports. Kept as a generic, archived automation pattern — adapt the profile and run on a schedule.

Local-first vector stroke recorder for scientific drawings. Designed to record stylus strokes as structured vector data rather than pixels, then replay/export high-resolution transparent timelapses for video editing workflows. Currently in early capture-core development.

Spark-based ingestion framework using NER-driven masking and de-identification for secure handling of clinical text and structured healthcare data.

Dagster-orchestrated ingestion system with automated validation, quarantine paths, profiling, and review workflows for unconfigured or anomalous data.

Interactive dashboards and data applications supporting pharmacy stakeholders with operational and analytical decision-making.

• M.S. in Computer Science — Old Dominion University (2024–2026)
• B.S. in Computer Science, Minor in Cyber Security — Old Dominion University (2017–2022)
• B.S. in Economics, Emphasis in Actuarial Math & Data Analysis — George Mason University (2009–2012)
• A.A. in Persian Studies — Defense Language Institute Foreign Language Center (2005–2006)
• Also studied music, intelligence operations, and general coursework at Berklee College of Music, Randolph-Macon College, Cochise College, and the University of Maryland Global Campus — an eclectic path from music school and military language training to economics, mathematics, statistics, computer science, data engineering, and applied AI.
• Research and project interests:
  • trustworthy AI infrastructure and agent verification
  • healthcare data systems
  • privacy-preserving ML
  • edge / local-first AI
  • biomedical LLM evaluation and calibration measurement
  • reproducible applied research workflows

I’m no longer focused on coursework. My attention now is on building a body of independent work: practical systems, research-informed prototypes, and technical writing that connect my healthcare data engineering background with the next generation of trustworthy AI tools.

• Agent verification, false completion, and goal-directed harnesses for data-engineering agents
• Model compression, quantization, calibration measurement, and inference efficiency
• Local-first and edge-ready AI deployment
• Agentic workflows for data engineering and research
• Hardware tinkering: Arduino, sensors, instrumentation, and small systems
• Security and privacy as design constraints, not afterthoughts

• Former Persian Linguist, U.S. Army 🇺🇸
• Lifelong reader; happiest in libraries and used bookstores 📚
• Drums, guitar, piano
• Pinball, live music, comedy, and learning things the hard way

Site: antonrasmussen.com

If you’re interested in:

• healthcare data platforms,
• trustworthy AI and verified agent systems,
• privacy-preserving systems,
• biomedical LLM evaluation,
• or building tools that people can actually understand and depend on,

feel free to explore my public repositories or reach out via the site.