Bill Baran.

Bill Baran

Software architect in Boise, Idaho, 28 years in.

I design and build systems that stay up for decades: government licensing and payment platforms, AWS infrastructure, and lately the plumbing that makes LLM agents dependable. Below are four apps I run every day, each with a working demo. There's also how I build with LLMs, and you can email me.

Things I run every day

These are self-hosted apps I use daily. Each demo runs the app's own page code on invented data, because the live data comes from services licensed for personal use.

Kotoba no Sekai

Learn a few real Japanese words a day from today's news.

TypeScript, Node, kuromoji, SQLite, Ollama, Docker

Equity Watch

Price alerts that don't stop at "it fired."

AI agents can drive it through WebMCP, the new browser standard for agent tools.

TypeScript, Node, WebMCP, MCP, Vanilla JS dashboard, Lambda

Outlier Caucus

Scores every stock trade Congress discloses for how unusual it is.

TypeScript, Node, MuPDF, Ollama vision OCR, SEC EDGAR, Docker

Spread Book

Daily exit decisions and greek risk heatmaps for a book of credit spreads.

JavaScript, Node, SheetJS, SVG, Docker, supercronic

Scheduled agents that are cheap and don't make mistakes

Three MCP servers I wrote, for Substack, Medium and Spotify, plus the Hermes skills and cron jobs that use them. They produce a daily reading digest and a playlist twice a week. Moving the bookkeeping out of the prompt and into code cut the Medium digest from $3.26 to $0.27 a run. The Substack digest and the playlist now run entirely on a local model.

Step through a real run
The model acts at steps 2, 3, 4 and 6. Everything in the server lane is code.
One digest run, split between the model and the server
Modeljudgment only: which posts, which stars, what each one saysServer (code)fetching, ranking, refs, layout, state1
digest_begin
Fetches everything new, drops what was already reported, ranks every headline with Jev, gives each post a ref: F1 T2 Y3
2
Shortlist
Picks the refs worth reading, starting from the top of the ranked list
3
Subagents read
On the local model, five refs each. They never see a URL
3b
read_post "F3"
Looks the ref up in the run file and fetches that exact post
4
Pick and summarize
Chooses the ⭐ posts and writes each gist, by ref
5
digest_finish
Checks every ref, saves state, then lays out the finished message
6
Reply
Sends the message unchanged. State is already saved

The model acts at steps 2, 3, 4 and 6. Everything in the server lane is code.

Skills

Software architecture
System integration, Event & queue-driven design, Static-first publishing, API design (REST, SOAP), Long-lived systems & migrations
AWS cloud
CloudFormation, S3 + CloudFront, Lambda, SQS, Route 53, IAM least-privilege, ECR, Bedrock
Docker, IaC & CI/CD
Docker & Compose, Infrastructure as code, GitHub Actions, Jenkins, Gerrit code review, Scheduled workers (supercronic)
Java
Spring / Spring Boot, Micronaut, Grails, Groovy, JUnit
TypeScript & Node
Node.js, TypeScript, Vitest / node:test, Playwright, Zod
LLM & agent engineering
MCP servers, Claude Code, Agent skills & cron workflows, Local models (Ollama, Qwen), Eval-driven prompt & classifier tuning
Data
SQL, MySQL, SQLite, RDBMS administration, PDF & OCR parsing pipelines
Security & compliance
PCI DSS, SOX, Web app penetration testing, PII handling, Encryption (incl. WebCrypto)
Payments & government systems
Payment processing services, Scheduled & recurring payments, Partner disbursements, Licensing systems

Experience

  1. 2000–now

    Senior Software Engineer, Access Idaho (Tyler Technologies)

    Boise, Idaho

    • Built ten government and licensing systems end to end, each with system integration, full accounting, and sensitive PII. Their business rules were subject to litigation and frequent rule changes.
    • Integrated with client systems through databases, binary file formats, batch FTP, and SOAP and REST web services.
    • Built payment processing services, including scheduled and recurring card and eCheck payments, and helped partners integrate with them.
    • Wrote billing reports and the daily deposit disbursements to partners.
    • Owned security scanning, deployments and escalated support. Helped implement PCI DSS and SOX compliance.
  2. 1998–2000

    DBA / Developer, Geneer

    • Database administration and application development.
  3. Education

    Rose-Hulman Institute of Technology

Smaller tools

Everything else on GitHub ↗
  • cloudformation-foundation ↗My personal infrastructure-as-code toolkit: reusable CloudFormation building blocks.
  • etf-compare ↗Upload ETF holdings files and compare their overlap side by side.
  • job-ledger ↗Bookkeeping for scheduled LLM jobs: runs, refs, seen sets and atomic state, so the model only judges. The shared library under the MCP servers.

Off the clock

Self-hosting

A home server runs my Docker stack: every app on this site, the scheduled jobs that publish them, an agent harness, and a local GPU model that handles most of the LLM work for free. If a SaaS has a monthly fee and an API, I have probably thought about replacing it.

3D printing

Functional prints for around the house and the homelab: brackets, enclosures, organizers and replacement parts. Measure, model, print, then reprint once more.

Learning Japanese

Slowly, through real news. Kotoba no Sekai began as my own study tool: a few new words a day, at my level, each with the sentence it came from. See Kotoba no Sekai.

Options and markets

I trade credit spreads and track a long-term portfolio, so I build my own tools for it: alerts that suggest what to do next, risk heatmaps, and a look at what Congress is trading. See Spread Book.

Contact

Hiring for architecture, AWS, Java, or agent and LLM engineering? Or curious how something here works? Email is the best way to reach me.

GitHub ↗

The address appears when you click, which keeps it out of spam scrapers.