Timeline

Aug 2026

Built "brain", an agentic second-brain system

Claude-operated personal knowledge system: a markdown + SQLite vault, custom skills (capture, meeting processing, briefings, standups, weekly review, career history, LinkedIn, site generation), a Next.js dashboard, and scheduled automation. 32 commits on the first day, through the job-search pipeline and the LinkedIn/brand layer. Runs entirely on the Claude Code CLI subscription — no API keys anywhere in the system.

Aug 2026

Rebuilt the portfolio site on Astro and made it publish itself from the vault

Second rebuild of milburnj.github.io, Next.js → Astro 5 with React islands. The larger change is that the site stopped being a place content is written: `scripts/portfolio_sync.py` regenerates its content from this vault, a scheduled task pushes, and GitHub Actions deploys. Approval lives in vault frontmatter — a confidentiality gate aborts the whole run before writing a file. 17 commits in August.

Jul 2026 – Aug 2026

Built plat, an agent-native hosting platform

Personal project (confirmed 2026-08-16): hosting for the small apps coding agents actually produce — too much work to deploy by hand, too small to justify a platform team. Contract-first app spec, an Azure control plane with Bicep infrastructure as code and Key Vault-backed configuration, a Fly.io workload plane, and three adapters over one service layer (REST API, typed client, MCP server) plus a CLI. Later added a web dashboard with email-link auth, GitHub import, and Alembic migrations. Playwright covers the auth/session path — and was proven by deliberately breaking the credentials/CORS path to see whether the suite caught it. It didn't, which is the finding.

Jun 2026 – Jul 2026

Rebuilt Pips' art pipeline around AI-generated soft-3D assets

Cross-platform fitness game (Expo/React Native, TypeScript, Supabase, Strava) where real workouts are the only way to grow a collection of 28 creature species — no mechanic ever sells creature power. This chapter pivoted the visual direction from pixel art to soft-3D: an image-to-3D asset pipeline (gpt-image-2 → Meshy) and a repeatable per-species styling seam. 43 commits in the repo from 2026-06-09; `projects/pips.md` records an earlier design start of 2025-11-29 (?) that no scanned artifact corroborates.

May 2026 – May 2026

Built paydirt, a problem-mining agent

Turns internet complaining into a short list of defensible business hypotheses. Adapters over Reddit, Hacker News, GitHub Issues and the App Store feed a two-pass pain-point extraction; candidates are embedded and clustered online, so recurring complaints collapse into one scored, velocity-tracked cluster instead of a thousand duplicates. A search stage mines complaints *about the incumbents*, then synthesis writes a hypothesis and an independent critic pass scores it — the guard against an LLM producing plausible generic slop forever. TypeScript, Next.js, Postgres + pgvector, Drizzle, Azure OpenAI.

Apr 2026 – Jun 2026

Built and shipped Julienne, an iOS cooking app, to the App Store

Receipt scanning (OCR plus a normalization layer that turns thermal-printer abbreviations into real ingredients) feeding a pantry, an assistant that suggests meals from what you actually have, StoreKit 2 subscriptions, Supabase sync, and CI for the app, the backend and the marketing site.

Feb 2026 – now

Production document-intelligence pipeline

Classification and field extraction over inbound clinical documents using Azure Document Intelligence custom models, Azure Functions, Blob Storage and PostgreSQL, with separate dev and production environments. 132 commits; August added a prod-measured cost analysis and a shadow-mode LLM extractor (see the 2026-08-17 milestone).

Dec 2025 – now

Operational analytics platform

Next.js/TypeScript reporting platform over a SQL reporting warehouse: provider utilization, billing performance, accounts receivable and provider productivity. Includes HR/payroll data reconciliation, role-scoped access control, and a golden-file regression workflow so report numbers can't silently drift. 142 commits and still active — August work centred on the billing-performance pages (one declared time window per page) and on verifying clinician-level figures against the source data.

Aug 2025 – now

Software Engineer, Aurora

Builds production ML pipelines and the data platforms and dashboards around them for a healthcare services organization. Confirmed by Jake 2026-08-16; earliest employer commit is 2025-08-29.

Jul 2025

Rebuilt the portfolio site on Next.js + Tailwind

Static-exported Next.js App Router site on GitHub Pages with an MDX blog pipeline; 18 project write-ups. Superseded by the Astro rebuild in 2026.

May 2025

M.S. Computer Science, University of Delaware

Coursework: computer vision, machine learning, large-scale machine learning, NLP, artificial intelligence, data mining, AI for bioengineering and bioinformatics, computer architecture, algorithm design and analysis, parallel computing, mobile robot programming, automata theory, cybersecurity, HCI.

May 2025

Artificial Intelligence Certificate, University of Delaware

May 2025

Evaluated RFdiffusion for constrained protein design

Unconditional backbone generation, symmetric oligomer design and hotspot-guided binder creation, scored through the ProteinMPNN → AlphaFold loop.

Feb 2025 – May 2025

Seafloor sediment segmentation with U-Net

Semantic segmentation of seafloor imagery plus bathymetry across 17 sediment and substrate classes; U-Net with ResNet backbones, class-weighted loss, IoU ≈ 0.74 on the held-out set.

Feb 2025

FathomNet scaling study

How detection models scale on the FathomNet underwater dataset — bounding-box coverage, class distribution and training feasibility across dataset sizes and taxa.

Feb 2025

BERT sentiment classifier on ~70k tweets

Fine-tuned BERT for three-class sentiment; ~92% accuracy on a curated ~70,000-tweet set.

Dec 2024

B.S. Computer Science (AI & Robotics concentration), University of Delaware

Nov 2024 – Dec 2024

Optimized matrix multiplication down the memory hierarchy

Successive optimization from naive loops through cache blocking to CUDA/cuBLAS, profiled with Nsight Compute against an OpenBLAS baseline.

Feb 2024 – May 2024

AI/ML Researcher, UD DeepREAL Lab

Led research on scalable multi-class classification of marine terrain with CNNs and U-Net/ResNet architectures; built preprocessing pipelines for sonar and bathymetric data and evaluated with IoU/Dice.

May 2023 – Jul 2023

Augmented Reality Engineer Intern, ZipCode Wilmington

Built a real-time AR learning app with WebAR and JavaScript — 3D rendering, object tracking, and marker-based interaction; improved latency and responsiveness.

May 2022 – Sep 2024

Software Engineer, IncNow (summers)

Designed, fine-tuned and deployed LLMs for a production AI customer-service assistant across web and Slack. Built RAG pipelines with vector-store retrieval and tuned embeddings for factual responses; ran prompt engineering and error analysis; maintained training logs and evaluation metrics for continuous deployment on AWS Lambda and SageMaker; added OCR/NLP document and form parsing.

Sep 2020 – May 2025

Student-athlete, University of Delaware Men's Crew

Five seasons alongside a full CS course load and research.