AI Agent Engineer · Full-Stack Developer — CS @ York University · Toronto, Canada

I turn AI agents from demos into production-grade systems — my favorite part is the unglamorous engineering that makes them reliable, observable, and affordable.

What I'm good at

• Agent orchestration — LangGraph state machines, typed tool contracts, bounded workflows that turn free-form agents into controllable pipelines
• LLM reliability — dual-layer guardrails (cut non-compliant output ~15% → <2% in production), fail-closed integrity gates, rule-engine-verified numbers, full audit trails
• Cost & performance — input-hash caching (60–70% hit rate, <50 ms), agent regression evals (~10% fix-pass lift, token cost ~2M → ~70k per task)
• 0→1 delivery — SwiftUI / Next.js / FastAPI / Spring Boot with CI/CD and monitoring; taking products from concept to production, solo or in small teams

What I'm into — agentic product systems · applied LLM infrastructure · AI + full-stack delivery · developer knowledge platforms

• 🏆 Adventure X 2026 — Track 1st Prize: Possibility, AI decision companion — solo build, 6 days, 189 commits, concept → shipped product
• 📜 Anthropic Claude Certified Architect — Foundations (CCA-F)
• 🔧 Merged PR in OpenClaw (385k★) — PR #2143: fixed model failover retrying cooled-down OAuth channels (worst case ~1 h user-facing stall → instant recovery)

1) SREDSimplify SR&ED claim console with evidence ingestion, structured pre-checks, and audit-friendly T661 drafting workflow for founders and reviewers.

2) Ai DIY Docs Engineering knowledge base covering CS core, AI agents, backend/frontend, and DevOps — built as a production-ready documentation platform.

3) Openbot Self-hosted GitHub maintenance bot — issue triage, bug reproduction, auto-fix, code review, Q&A. Bounded workflows + regression eval: ~10% fix pass improvement, token cost ~2M → ~70k per task.

🏆 Possibility — Adventure X 2026 Track 1st Prize AI decision companion for young professionals navigating career transitions. A 3-stage funnel: streaming AI chat surfaces emotional context & decision signals → structured matching with 3 real human experiences across different outcomes with rationale → paid conversion. Built solo in 6 days (189 commits) from concept to shipped product. 🏆	🥈 SplitBase — NTU x Base Web3 Hackathon 2nd Place Base Pay bill-splitting app for group settlements with USDC, real-time status tracking, share links/QR flows, and optional NFT receipts.
SmartDiet Copilot Agentic AI dietitian that combines receipt/fridge/meal perception with goal-aware planning to reduce waste and deliver actionable meal decisions.	FitLogic AI fitness coaching experience focused on adaptive training + nutrition logic, progress analytics, and coach-assisted personalization.

Turning AI agents into production systems with measurable outcomes.