Structured AI · Founding Engineer.
Fourteen weeks of agent infrastructure, sandboxed runtimes, an iOS app, and durable chat — all in production.
Joined at the founding-engineering stage with a clear mandate: ship the agent infrastructure, the iOS client, and the deploy pipeline that production needed. Most of the engineering decisions had compound constraints — enterprise compliance, cross-platform parity, deploy speed — that rewarded engineers who could move fast across the stack.
Fourteen weeks of velocity: 1,035 commits, ~74/week median, peak week of 145. The work spanned backend (FastAPI + Postgres + Redis + RPC bridge), agent loop (graph runtime + LLM with on-demand tool loading), code execution (in-house sandbox built for Zero Data Retention compliance), iOS / iPadOS client from zero, and CI work that materially cut staging deploy time.
Most architectural detail beyond this public-pattern summary is NDA-bound. The deeper version is available in 1:1 conversation under appropriate scope.
In-house code-execution sandbox for ZDR
Enterprise clients required Zero Data Retention, and hosted tool-execution backends typically don't satisfy ZDR. The shape: a request-interception layer that handles tool-execution traffic locally — untrusted code runs in a hardened in-house pod (JSON-RPC over child subprocess, AST validation, bubblewrap + seccomp); customer data wipes immediately and never reaches the model provider.
Multi-vendor agent (reasoning LLM + delegated vision model)
Reasoning LLM drives the loop; a separate vision model is delegated for high-resolution document inspection through a multi-tile rendering pipeline that yields significantly higher effective resolution than naive full-page rendering. Different providers, different strengths — the agent uses each where it wins.
Synthesis over framework adoption
Built three prior agents on different framework choices. None hit the bar. The shipped agent is a synthesis: takes the strongest pattern from each (prefetch-backed tools and server-side compaction, workspace + execute model, minimal-tool philosophy, adaptive thinking) and strips out every line of framework overhead.
Modular pipeline with on-demand tool loading and persistent skill / memory injection. Cache-friendly via ephemeral prompt-cache primitives — the right shape for cost-sensitive long-context agents.
Domain documents (large-format drawings) lose critical detail under naive full-page rendering. Multi-tile rendering yields significantly higher effective resolution at the same per-tile cost. Multi-vendor design where each provider plays its strength.
Linux user-namespace isolation + syscall filtering + pre-execution AST static analysis. Same-class hardening as production code-execution platforms (Replit / E2B), built in-house for ZDR-compliant enterprise deployment.
Hosted programmatic tool-calling backends don't satisfy Zero Data Retention. A custom interception layer keeps execution local while preserving the model's code-quality. ZDR compliance without sacrificing capability.
Cross-process typed RPC at scale. Schemas double as the API contract — no drift between client and server.
Chat survives network blips, pod restarts, and deploys without lost messages. Replay-on-reconnect is the modern primitive for production conversational UI.
Container-native serverless compute for the agent pod, registry-side image builds with cached layers, and CPU-only image trimming. The net effect was a materially faster staging cycle.
- →Founding-engineering velocity isn't typing speed; it's the ability to make 'good enough now, refactor next sprint' calls without stopping the line. Every week I shipped, I also paid down a known piece of debt I'd taken the prior week. Net: forward motion every Monday.
- →Multi-vendor agent design is increasingly important — single-provider lock-in is a liability for both compliance and capability. The pattern of 'reasoning model + delegated vision model' will keep generalizing.
- →The most valuable infrastructure I shipped was the CI work. A materially faster staging cycle changed how the whole team iterated.