Skip to main content
jeff.leung
Reliability narrative portfolio

I build engineering systems that make AI-assisted delivery reviewable without burning tokens.

Senior software engineer focused on the substrate around coding agents — plans, scoped tools, verification gates, and sanitized receipts. Three case studies follow.

View case studies Book an open conversation
  • Plan

    Intent becomes a reviewable contract.

  • Verify

    Outputs pass explicit gates before completion.

  • Receipt

    Work leaves an audit trail that can be sanitized.

Reliability thesis

AI-assisted delivery becomes reviewable when the engineering system around it is explicit.

Coding agents are fast at producing output. They are slow at producing reviewable output. The difference is not the model; it is the engineering system around the model — the plans, scoped tools, verification gates, and receipts that turn agent activity into something a senior engineer can defend in review.

  • 01

    Plans are written before execution. Scope, non-goals, and confidence are explicit.

  • 02

    Agents act through scoped tools whose calls and outputs an engineer can audit.

  • 03

    Repeatable work runs hands-off through CLI recipes and gates, so automation reaches exactly as far as the audit trail does.

  • 04

    Scoped, economical context keeps agents fast and reviewable — token economy is a reliability property, not a cost trick.

  • 05

    Work is only complete after verification has run and its output is recorded.

Case studies

A path through the reliability narrative.

Three sanitized systems, shown as proof cards first and deep technical narratives second.

  1. Case study 01

    Small agents. Deterministic gates. Reviewable delivery.

    Instead of loading every tool and rule into one agent, Pi routes repeatable work through CLI recipes, gives focused agents only the context they need, and requires isolated validation before QA handoff.

    AI engineering Backend / Platform Verification
  2. Case study 02

    Agent Tools Workstation System

    AGENT-TOOLS keeps skills, commands, sub-agents, and host policy in one git-backed control plane, so Codex, Claude Code, and OpenCode can share the same operating model across machines.

    AI engineering DX Infrastructure
  3. Case study 03

    MCP Router

    One shared local MCP router instance for multiple coding-agent hosts, with cached capability lookup, isolated plugin runtimes, and measurable routed-vs-direct overhead.

    Backend / Platform Infrastructure AI engineering
Capability matrix

Five proof areas, one reliability thesis.

AI, backend, frontend, infrastructure, and security are not separate pillars in this portfolio. Each is presented as evidence the same thesis can hold under production constraints.

AI-assisted engineering
Designs plan-first workflows, scoped agent tools, sub-agent personas, and host-portable conventions for coding-agent harnesses.
Tie back · Makes AI-assisted work auditable so engineering leaders can ship it without sacrificing review authority.
Backend / Platform
Builds typed APIs, plugin runtimes, and toolkit-style platforms with explicit contracts and lifecycle hygiene.
Tie back · Provides the substrate AI tools route through — without typed surfaces, reliability claims do not survive contact with production.
Frontend craft
Ships static-first, accessible, motion-aware interfaces with strict design tokens and reduced-motion fallbacks.
Tie back · Turns reliability evidence into something a hiring manager can read in 30 seconds and an engineer can read in five minutes.
Infrastructure
Reproducible workstation setup, strict-idempotent installers, doctor flows, and static-first deploy pipelines.
Tie back · A reliability story written on one configured laptop is not a reliability story — reproducibility is what makes it portable.
Security
Sanitization checklists, privacy guards, capability allow-lists, and CSP-tight static deploys for portfolio-style surfaces.
Tie back · Reliability is a security property — it means callers can trust both the output and the disclosure boundary it ships under.
Get in touch

Pick a path that fits the conversation.

Start with the case studies for the long read, or book an open conversation for a 30-minute technical chat.

View case studies Book an open conversation
Also at linkedin.com/in/jeff-leung-lab github.com/WonderJL