Archive

Experience

Four jobs, two highlights each. Open More for the rest and the writeups.

2026 — present

Chiang Rai, remote

College Professor

HAIT

IT, the public site, and the first AI class.

  • Run hosting and security for the school.
  • Rebuilt the public site and the PHP underneath it.
More2 more highlights · 4 writeups
  • Wrote the school’s first AI curriculum and taught the first cohort.
  • Set up OpenClaw and OpenAgent workflows, and started a data-recollection project with a collaborator at Korea University.
Public siteA rebuilt public site that loads cleanly.
PHPHosting
  • Rebuilt the front end and upgraded PHP.
  • Fixed the hosting config so pages load reliably.
Problem
The school needed a public site that loaded cleanly.
Constraint
They wanted to keep their existing host.
Decision
I redesigned the front end, upgraded PHP, and fixed the hosting config.
Outcome
The site is live at haitedu.com.
Institutional ITHosting and security for the school.
HostingSecurity
  • I handle hosting and security day to day.
  • The public site stays up.
Problem
The school needed someone on hosting and security.
Constraint
The public site was going live at the same time.
Decision
I took over hosting and security.
Outcome
Both are still covered.
AI curriculumWrote the first AI course and taught it.
  • Wrote the school’s first AI course.
  • Checked the lab machines, then taught the class.
Problem
They needed a first AI course and someone to teach it.
Constraint
It was the first cohort, and the lab machines needed checking first.
Decision
I wrote the course, checked the machines, and taught the class.
Outcome
The first students finished the course.
Multi-agent orchestrationOpenClaw and OpenAgent for the school, plus a recollection project with a Korea University collaborator.
OpenClawOpenAgent
  • OpenClaw and OpenAgent, set up so students and staff can run them.
  • A separate recollection workflow with a collaborator at Korea University.
Problem
The school wanted agents students and staff could run, plus a data project with someone at Korea University.
Constraint
Everything had to run on the lab machines.
Decision
OpenClaw and OpenAgent for teaching, and a separate recollection workflow with that collaborator.
Outcome
Students and staff use the teaching agents. The recollection project is planned.

Sep 2025 — Aug 2026

New York

Founding CTO

Stealth fintech

Kraken futures, ingest under 100ms.

Read the write-up

  • Built a trading engine on Kraken crypto futures in Python and Kafka, with under 100ms from feed to book.
  • Wrote agents that check portfolios against the SQL ledger.
More1 more highlight · 4 writeups
  • Owned the models: PyTorch time-series models that retrain around the clock, news processors that catch sudden moves, and a daily pre-market brief that cut morning analysis by 90%.
Kraken futures ingestionPython and Kafka on the Kraken feed, under 100ms.
PythonKafka
  • Millions of ticks, under 100ms end to end.
  • Python and Kafka on the official feed.
Problem
They needed a live feed going into the trading book.
Constraint
Kraken crypto futures, and under 100ms.
Decision
Python and Kafka on the official feed, with agents on top of the book.
Outcome
Ticks come in and orders go out.
Forecasting and retrainingPyTorch models that retrain around the clock.
PyTorch
  • Time-series models for live signals and backtests.
  • A retraining job that runs all day, plus local training on a Mac.
Problem
They wanted signals that kept up with the market.
Constraint
Training had to continue throughout the day.
Decision
PyTorch time-series models that retrain all day.
Outcome
Live signals and backtests, with models that retrain on their own.
News processorsNews text in, pivot flags out.
  • It runs on the same schedule as the models.
  • It reads live headlines and flags pivots.
Problem
Some moves show up in the news before they show up in prices.
Constraint
The input is live headlines.
Decision
A news processor running on the same schedule as the forecasting job.
Outcome
Pivot flags from headlines, alongside the time-series signals.
Portfolio audit and pre-market briefAgents on the SQL ledger, and a morning brief that cut the desk’s analysis work by about 90%.
PythonSQLLLMs
  • Agents that query the SQL and check portfolios.
  • A daily pre-market brief that cut the morning analysis by 90%.
Problem
The morning portfolio check was all manual SQL.
Constraint
The models had to work with the ledger they already used.
Decision
Agents that query the SQL, plus a PyTorch-based brief before the open.
Outcome
They run the brief every morning.

May 2022 — Jul 2025

New York

Full Stack Engineer

Goldman Sachs

80% faster processing, embedded with the team, 50+ internal teams.

Read the write-up

  • Embedded engineer between developers, traders, and lawyers, building one reporting path so the numbers stayed auditable and compliant.
  • Designed and shipped a beneficiary management system for 10,000+ client accounts and cut data processing time by 80%.
More3 more highlights · 4 writeups
  • Scaled Java and Kafka pipelines and REST APIs used by 50+ internal teams, cutting data-access latency by 15%.
  • Built a Selenium and TestNG suite that raised test accuracy 22% across hundreds of financial and compliance scenarios.
  • Kept supporting the trading, compliance, and market-data APIs after they shipped.
Beneficiary management10,000+ accounts, with processing time down 80%.
JavaKafka
  • Designed and shipped the beneficiary system behind 10,000+ client accounts.
  • Cut data processing time by 80%, and stayed on it after launch.
Problem
More than 10,000 accounts needed a beneficiary system that people could actually use.
Constraint
Compliance requirements, and more than 50 teams reading the same data.
Decision
Java and Kafka, with a REST API for accounts.
Outcome
Processing time dropped 80%.
Java, Kafka, and RESTPipelines and APIs used by 50+ internal teams, with latency down 15%.
JavaKafkaREST
  • Scaled Java and Kafka pipelines and REST APIs used by 50+ internal teams.
  • Cut data-access latency by 15%.
Problem
Internal teams needed to share account data over an API.
Constraint
More than 50 teams depended on it, so the APIs had to stay up.
Decision
The same Java and Kafka pipelines, exposed over REST at /v1/accounts.
Outcome
Access latency dropped about 15%.
Forward deploymentDevelopers, traders, and lawyers on one reporting path.
  • I sat between the developers, traders, and lawyers who all used the same reporting path.
  • The goal was one set of numbers all three groups would sign off on.
Problem
Developers, traders, and lawyers all had to sign off on the same numbers.
Constraint
Three groups, one report.
Decision
I worked between the three groups and built one path.
Outcome
One set of numbers all three groups signed off on.
Selenium and TestNGTest accuracy up 22%.
SeleniumTestNG
  • Built a Selenium and TestNG suite across hundreds of financial and compliance scenarios.
  • Raised test accuracy 22%.
Problem
KYC and compliance checks needed a proper test suite.
Constraint
Hundreds of cases, and the APIs were already in production.
Decision
Selenium and TestNG across those cases.
Outcome
Test accuracy went up 22%.

Oct 2020 — May 2022

La Brea, CA

Full Stack Software Engineer

One Dave Software

Curtis inventory, defense apps, email 70% faster.

Read the write-up

  • Built inventory pipelines and frontends for Curtis, a government nuclear pipeline contractor.
  • Led a system-wide upgrade that cut email-server response time by 70%.
More1 more highlight · 3 writeups
  • Built and maintained secure National Defense web applications.
Curtis inventoryInventory tracking for Curtis, a government nuclear contractor.
  • The pipelines, and the screens on top of them.
  • Built inside their existing contractor systems.
Problem
Curtis needed inventory tracking for a government nuclear contract.
Constraint
We had to work inside their existing systems.
Decision
I built the pipelines and the frontends.
Outcome
They use it to track inventory.
Email systemsA system-wide upgrade that cut response time by 70%.
  • Managed email for the same clients.
  • Upgraded the whole mail path, which cut server response time by 70%.
Problem
The same clients needed faster email.
Constraint
The upgrade had to cover the whole mail path.
Decision
I upgraded the mail servers.
Outcome
Response time dropped 70%.
National Defense appsWeb apps for national defense clients, with security built in from the start.
  • Built and maintained full-stack apps for national defense clients.
  • Security review started with the first ticket.
Problem
Defense clients needed web apps that stayed up and stayed locked down.
Constraint
Security review from the first ticket.
Decision
I built the apps and kept them running.
Outcome
They stayed in service.