From forecasts to agents in production.
I build the platforms that let AI make real decisions — from million-series forecasting engines to production agent systems.
02 / About · Now
Engineer-turned-builder of decision intelligence.
Nine years building enterprise decision-intelligence software. I started on distributed forecasting infrastructure — Kafka, Dask, Kubernetes, time-series ML — and grew into leading an agentic AI platform end to end: assistant runtime, workflow engine, memory, retrieval, LLM gateway, guardrails and evals. I still write code every day. I care about systems that are measurable, safe, and boring to operate.
- Role
- Director, AI/ML Engineering at Aera Technology — hybrid IC + engineering leader
- Orbit
- ~9 years at Aera, 2017 → now
- Focus
- Agentic AI platforms, with roots in supply-chain forecasting
- Long game
- Technical leadership of an early-stage AI product. Still in the code.
“Claude Code, configured well, is a workbench — not a chatbot.”
“A skill borrows your context; an agent creates a new one.”
“You can't optimize what you don't measure.”
“Build first. The meetings will come.”
“Trust structured artifacts, not LLM narration.”
03 / Orbit · Work eras
Seven eras, one system.
Scroll or click a planet
2017
04 / Systems
Systems I've built.
Two orbits: the forecasting and ML platform I grew up on, and the agentic platform I lead now.
Inner orbit · Forecasting & ML
Outer orbit · Agentic platform
05 / AI · Agentic engineering
Workbench, not chatbot.
Claude Code daily across ~40 projects — worktree-per-PR, persistent memory, skills, subagents, hooks, session-cost analysis.
Prompt-cache hit rate on a 5M-token session.
PRs authored in one review cycle.
PRs reviewed in the same cycle, across ~25 repos.
A multi-system ops lookup, now a slash command.
Practice
Measured
88% prompt-cache hit rate on a 5M-token session; turned token data into team habits.
Ops lookups
A five-minute multi-system lookup turned into a four-second slash command; wrote the playbook so the team could repeat it.
Wiki with a freshness check
Answers tied to source files, flagged when code changes after the doc was written.
AI-native repository spec
Adopted in multiple repos; internal talk: AI-First Engineering.
Personal lab · moons
“A skill borrows your context; an agent creates a new one.”
06 / Tech stack
Everything in orbit.
Hover the belt to pause
07 / Writing · Notes
Notes from the workbench.
Reading shelfDDIA · Fundamentals of Software Architecture · Think Like a CTO · Learning Ray · Modern Time Series Forecasting · TimesFM / TimeGPT / DeepAR papers · Attention Is All You Need