Engineering with AI. Shipping with Precision.
Building production systems with AI-assisted workflows. The right tool for each phase. Human judgment at every checkpoint.
How I Work with AI
A deliberate workflow where AI amplifies engineering judgment, not replaces it. Every capability is practiced, not just listed. The human stays in the loop at every decision point.
Agent Orchestration
orchestrationCoordinating multi-agent systems, delegating tasks, and synthesizing results across parallel AI workflows
- Multi-agent pipeline design and coordination
- Task decomposition and parallel delegation
- Agent memory, tool-use, and context management
Context Engineering
engineeringStructuring environment, data, and interaction flows so AI systems produce reliable, grounded outputs
- System prompt architecture and optimization
- RAG pipeline design with vector retrieval
- MCP integration for external data sources
AI-Assisted Architecture
engineeringUsing AI as a thinking partner for system design, tradeoff analysis, and large-scale refactoring
- AI-driven system design and planning
- Codebase-wide refactors with full context
- Automated scaffolding and boilerplate generation
Harness Engineering
qualityBuilding the runtime layer around AI agents: hooks, evaluation, safety controls, and CI/CD integration
- Pre/post-action hooks and quality gates
- AI-driven code review and PR analysis
- Automated testing, linting, and deployment pipelines
Prompt Engineering
engineeringDesigning and iterating on prompts for complex multi-step workflows with measurable quality benchmarks
- Chain-of-thought and few-shot prompt design
- Prompt evaluation and regression testing
- Domain-specific prompt libraries and templates
Agentic Workflow Design
orchestrationDesigning end-to-end development workflows where AI handles execution and humans own decisions
- Research-plan-execute development cycles
- Delegate-review-own operating models
- Critical evaluation of AI-generated output
Development Pipeline
Architect
AI-assisted system design and planning
Explore tradeoffs, generate architecture plans, and reason about system design with AI as a thinking partner before writing code.
Build
Agent-orchestrated implementation
Decompose features into agent-ready tasks, delegate to parallel AI workflows, and synthesize results with human oversight at every merge point.
Validate
Automated quality and critical review
AI-powered PR review, automated lint and type-check hooks, regression testing pipelines. Final judgment is always human.
Ship
CI/CD with AI-driven confidence
Automated build verification, static analysis gates, and deployment pipelines with AI-assisted monitoring and rollback.
Architect
AI-assisted system design and planning
Explore tradeoffs, generate architecture plans, and reason about system design with AI as a thinking partner before writing code.
Build
Agent-orchestrated implementation
Decompose features into agent-ready tasks, delegate to parallel AI workflows, and synthesize results with human oversight at every merge point.
Validate
Automated quality and critical review
AI-powered PR review, automated lint and type-check hooks, regression testing pipelines. Final judgment is always human.
Ship
CI/CD with AI-driven confidence
Automated build verification, static analysis gates, and deployment pipelines with AI-assisted monitoring and rollback.
Tech I Use
The working toolkit behind the workflow above.
My Projects
A collection of tools and platforms built to solve real-world problems in automation, AI, and software engineering.
AI-Native Portfolio
This site. Built entirely with AI-assisted workflows to demonstrate modern development practices. Architecture planned by coding agents, implemented with AI-augmented editors.
Meta-demonstration: built BY the AI workflow it showcases
liveOmabite
A wizard-driven dish picker for UberEats. Five short steps land you on one specific dish at one specific restaurant, with reasoned "why this" and a deep link to order. A single SSE pipeline orchestrates Brave search, Apify menu scraping, and Kimi ranking, then validates the pick against the menu and the user's allergens.
Streamed multi-stage pipeline: search → shortlist → menu → dish pick, with graceful degradation
Experience
What I'm shipping right now. Full-stack delivery with deliberate AI leverage at every checkpoint.
- 01
Architect AI-assisted development harnesses, custom agent profiles, MCP tool integrations, and phase-gated workflows that keep LLM output aligned with requirements instead of drifting into confident hallucination.
- 02
Turn vague product asks into typed specs and testable contracts before code gets written. Result: fewer iterations, cleaner PRs, predictable delivery.
- 03
Maintain a cross-session knowledge base that feeds agents hard-won debugging patterns (streaming protocols, effect races, LLM output validation, cache invariants), so AI tools learn from past mistakes instead of re-solving them.
- 04
Ship LLM-powered product features end-to-end: custom chatbots, RAG pipelines over domain data, streaming UX with completion-vs-truncation protocols, and output-contract validation that surfaces drift as warnings rather than silent corrections.
- 05
Build SvelteKit apps end-to-end for thousands of students: Svelte 5 rune state models, SSR, Azure AD JWT + LDAP auth, accessible mobile-first UI, MSSQL stored procedures behind pooled connection wrappers, Docker + Azure Pipelines CI/CD across dev, QA, and prod.
- 06
Go-to teammate for AI tooling adoption and for debugging non-obvious concurrency, streaming, and LLM-boundary failures.