I build developer-facing systems that make complex workflows easier to integrate, debug, and operate. My work connects API contracts, SDK behavior, failure recovery, agent permissions, and the evidence behind a technical decision.
A system should explain what happened, why an action was allowed or blocked, how to recover, and which assumptions shape its value.
Portfolio · Engineering evidence · LinkedIn · Medium · Contact
SignalOS — The Developer Experience Reliability Platform
One payment-launch investigation connects contract findings, an outdated quickstart, timeout recovery, duplicate protection, agent approval, policy replay, and an exportable evidence trail.
- Analyze OpenAPI contract gaps and supported breaking changes.
- Reproduce local HTTP failures; retry a committed operation without duplicating it.
- Inspect HMAC verification, replay windows, and synthetic event ordering.
- Require an expiring, argument-and-policy-bound approval before a simulated financial action.
- Preserve investigation history in SQLite and export Markdown, JSON, HTML, or CSV.
- Inspect estimated economics with explicit inputs; unknown costs remain unknown.
Working local MVP: React, TypeScript, Node 24, SQLite, 44 behavioral tests, desktop/mobile browser checks, and GitHub Actions. The broader platform is a documented roadmap; the current SDK, documentation, and economics features support the payment scenario.
Case study and console preview · Architecture · Engineering article · Verification
| Project | Purpose | Implementation to inspect |
|---|---|---|
| AgentLedger | AI Agent Economics and Decision Ledger | Model/tool/retry/review cost estimates, persistent events, unknown-cost propagation, human baseline and risk-adjusted value |
| TrustBoundary | AI Agent Permissions, Policy and Human Approval | Strict tool schemas, role/scope gates, expiring approvals, idempotent execution and policy-aware replay |
| EvidenceGraph | Research Provenance and Decision Support | Source-to-claim links, contradictory evidence, freshness, human decisions and preserved evidence snapshots |
| WorkflowROI | Automation Economics and Human-in-the-Loop Simulation | Manual/partial/AI-assisted comparison, review and error costs, adoption, break-even and sensitivity analysis |
| DeveloperJourney Observatory | Developer Onboarding, Friction and Recovery Analytics | Synthetic journeys, first-success and recovery metrics, privacy validation and documentation comparisons |
These five applications use React, Node 24 and SQLite, with local setup, tests, browser verification, reports, architecture notes and engineering articles. Explore the collection.
| Project | Purpose | Implementation to inspect |
|---|---|---|
| DX Orbit | Developer Experience Intelligence for OpenAPI | Explainable contract findings, integration friction, revision comparison and report export |
| DevStart | API Developer Onboarding & Recovery Environment | Authentication, validation, rate limits, failure states and actionable recovery scenarios |
| HookForge | Webhook Reliability & Debugging Workbench | Signatures, retries, duplicate delivery, event ordering, malformed payloads and recovery |
| ToolTrust | AI Agent Tool Reliability Infrastructure | Tool schemas, permissions, side effects, confirmation, idempotency and replayable simulated execution |
| SignalDesk | Developer Feedback Intelligence | Structured friction and documentation feedback, prioritization, resolution and issue-draft export |
| HookForge SDK | TypeScript SDK & Generator Evaluation | MIT SDK generated with Voxgig, setup/examples, generated checks, local HTTP verification and a developer-experience report |
| Project | Focus |
|---|---|
| ForgeFlow AI Automation | React/TypeScript and FastAPI workflows, validated inputs, SQLite run history, visible fallback behavior and human review |
| MailTrace DX Lab | Signed webhooks, retries, idempotency, event ordering and developer-facing failure investigation |
| ClearRoute API | Typed validation, legal workflow transitions, predictable errors, audit events, tests and CI |
| AccessPath Console | Keyboard-first workflows, semantic interfaces, visible focus, accessible recovery and accessibility checks |
- Technical Portfolio / Signal Engine: a React/TypeScript review experience connecting project narratives, actual application previews, public source, tests and implementation boundaries. Visit the website.
- Portfolio Assets: the supporting visual asset repository for the portfolio.
- Follow the whole integration: start with SignalOS's payment investigation and compare its report with the source and tests.
- Inspect one boundary deeply: TrustBoundary for approval/idempotency; HookForge for delivery/recovery; DX Orbit for contract findings.
- Challenge a decision: change an assumption in AgentLedger or WorkflowROI, or inspect conflicting evidence in EvidenceGraph.
- Make inputs, state transitions and failure paths explicit.
- Validate before acting; keep permission and approval decisions inspectable.
- Test recovery, malformed requests and duplicates alongside the happy path.
- Keep measured behavior, estimates, assumptions and synthetic fixtures distinguishable.
- Document what runs today and what production operation would still require.
Working stack: TypeScript · React · Node.js · Python · FastAPI · Pydantic · Zod · REST/OpenAPI · JSON Schema · SQLite · pytest · Vitest · Playwright · Docker · GitHub Actions
Evidence boundary: These are independent engineering projects. Simulated payments, agent execution, feedback and onboarding sessions are labelled. Forecasts are not measured savings; tests and local verification do not imply production adoption, scale, authenticated security or customer outcomes. Each repository documents its own implemented scope and limitations.