Solutions architect building agentic engineering workflows — I run OpenAI Codex and Claude Code side-by-side, in parity, on real production codebases, and I've spent 20+ years deploying enterprise software into regulated industries (pharma, financial services, government, oil & gas).
Currently: Senior Principal Architect in OpenText's Customer Success Organization, where I embed with enterprise engineering teams on cloud, GenAI, and content-platform programs — increasingly with coding agents as first-class actors in delivery.
| Project | What it is |
|---|---|
| gaurava | Shipped App Store health & medication tracker (watchOS + widgets, Hindi/Tamil/Telugu) — full source of a production iOS app built end-to-end with Codex (incl. the Build iOS Apps plugin) and Claude Code |
| withful | Shipped App Store family-moments journal, live in 175 countries (en/fr/zh) — SwiftData + CloudKit private sync, 5 widget surfaces, App Intents/Siri, agent-governed delivery |
| intelli-expense | On-device receipt scanning for iPhone + Mac (Apple Foundation Models) with a review-gated agent import bridge for Codex and Claude Code |
| stepback | Hands-free workout routine builder & player (iPad/iPhone/Mac) with an agent bridge — ask your agent for a training week, it lands in the app, review-gated |
| underwriting-workbench | Evidence-first life-insurance underwriting assistant POC — deterministic plan-aware rules engine, citation-grounded hybrid RAG (LanceDB + bge-m3), MLX LoRA fine-tune with MLflow-tracked bake-off + divergence guard, FastAPI review workbench; fully synthetic fictional insurer |
| manifest-overlay-studio | Enterprise Documentum overlay generator built for agents to operate safely — one validated Pydantic manifest → a byte-stable Helm overlay tree; CLI/REST/web façades held in bijection by test; 1,100+ tests incl. byte-parity golden fixtures; ships as a single-file .exe for locked-down jump hosts |
| goal-campaigns | Pattern for multi-day autonomous campaigns with Codex goals — adversarial convergence gates, state-file governance, and a demo API + k6 harness |
| hanuman-dev | Multi-model plan→work→review orchestrator with dedicated Codex and Claude Code adapters |
| openai-agents | Personal AI companion on the OpenAI Agents SDK (TypeScript) with deny-by-default tool policy gating |
| workspace-bootstrap | Agent-governance scaffold: AGENTS.md/CLAUDE.md contracts, lifecycle hooks, session journals — for Codex and Claude Code |
| doot-companion | Sanitized 215K-LoC personal AI companion snapshot with governed Codex/Claude Code contributors and 271 verified tests |
| session-journal | Installable continuity discipline with shared Claude Code/Codex hooks, agent-led adoption, secrets guard, and cross-platform tests |
| simpalarm | Shipped macOS menu-bar alarm app, distributed via its own Homebrew tap |
| project-context | Pattern for persistent, context-rich AI agent workspaces |
Pattern: the agent write bridge. Intelli-Expense and StepBack both ship the same pattern: a local folder-drop protocol that lets a coding agent (Codex / Claude Code) author in-app data — receipts, workouts, routines, weekly plans — behind explicit human review. The agent reads the app's manifest, extracts fields conversationally, and drops JSON commands into an app-owned inbox; the running app validates, persists, and syncs. Agents never touch the data store, and the protocol has no delete capability. Protocol docs: intelli-expense/plugin · stepback/plugin.
Two consumer apps on the App Store — Gaurava (watchOS + widgets, localized in Hindi/Tamil/Telugu) and withful: family moments (available in 175 countries) — built end-to-end with agent workflows, including Codex's Build iOS Apps plugin — and both are now fully open source (see the table above). Most of my day-job work (enterprise Documentum/cloud deployments, agentic ops tooling, LLM fine-tuning POCs) lives in private repos — but the architecture of the deployment tooling is public as manifest-overlay-studio, re-cut against a fully fictional customer.
Also active as dctmfoo123.