Agent roles & boundaries
I turn product requirements into precise agent contracts: responsibility, tone, tools, refusal behavior, handoffs and explicit expertise boundaries.
Senior AI systems engineer · Production prompt engineer
I turn AI behavior into production software.
I independently architect agent systems, prompts, tools, context, evaluations, backends and product surfaces. My strongest discipline is production prompt engineering: making models behave reliably inside real workflows, not optimizing isolated chat responses.
57
AI modules in one production platform
18
specialized agents in one system
100+
voice intents designed and implemented
12
products live on the App Store
Flagship systems
The scale was already there. Now it is inspectable.
These are not tutorial builds or UI concepts. I independently designed and implemented the systems below. The repositories remain private; each case study exposes the architecture, engineering decisions and verifiable scope without exposing confidential source code.
Prompt engineering
Prompts are part of the architecture.
Building dozens of agents, voice workflows and AI products taught me where prompting actually succeeds or fails: at the boundaries between context, tools, policy, evaluation and product behavior.
Context engineering
I decide what the model sees, when it sees it and how memory, retrieval, user state and tool results are compressed into useful context.
Tool-use prompting
Schemas, tool descriptions, confirmation policies and recovery instructions are designed together so agents act predictably, not just speak convincingly.
Eval-driven iteration
Prompts are versioned against failure cases, semantic graders, A/B evaluations and rollout gates. Quality is measured as behavior, not writing style.
Guardrails & escalation
Confidence, risk and policy thresholds decide when the model may proceed, ask a clarifying question, fall back or route the decision to a human.
Cost-aware model routing
Prompt design and provider selection are optimized together, reserving expensive reasoning for tasks where it materially improves the outcome.
How I operate
AI speed with senior accountability.
01
Own the architecture
I define the data model, trust boundaries, agent contracts, integration surface, failure modes and deployment path before complexity compounds.
02
Build AI-native
Claude Code and Codex are daily engineering tools. Architectural judgment, product decisions, validation and final accountability remain mine.
03
Gate production quality
Type checks, automated tests, agent evaluations, cost and latency controls, security review and observability are part of the system itself.
04
Ship the whole product
I move across frontend, backend, AI, database, mobile, payments, integrations and deployment without handoff gaps.
Working stack
Production tools, not buzzwords.
Next.js 16React 19TypeScriptPythonFastAPISupabasePostgreSQLClaude Agent SDKAnthropicOpenAIGeminiRAGEmbeddingsVercelAWSFirebaseSwiftUIReact NativeKotlin
Need one engineer who can own the entire AI build?
I turn ambiguous requirements into architecture and carry the work through agent behavior, application code, integrations, testing and production deployment.