AI practitioner. Still shipping code.
I started MTechZilla because too many good ideas were dying from bad execution. The highest-leverage work has since moved from building the app to building the AI system inside it.
That's where my own hands are now - retrieval, agents, and evals - on top of a decade of full-stack, mobile, and IoT work that taught me what production actually costs.
I write about this on freeCodeCamp →Users
Product
AI System
- Retrievalgrounded in your data
- Agentstools, plans, retries
- Evalsproof it still works
Your data
I write the eval suite before I write the demo.
Every agent gets a cost budget and a rollback plan.
If it only works on the happy path, it isn't done.
The stack, layer by layer.
Retrieval sits at the base, agents build on it, voice is one interface onto it - and evaluation and infrastructure wrap the whole thing. Five of these go deep enough to deserve their own guide.
Five deep dives
Grounding model outputs in your own data instead of hoping the weights remember it.
Read the guide →LLM EvaluationMeasuring whether a change made the system better, not just different.
Read the guide →MCPThe emerging standard for wiring tools and data sources into any model client.
Read the guide →AI AgentsSystems that plan, call tools, and act - with real error handling, not a happy-path demo.
Read the guide →Voice AI AgentsAgents that hold a real-time conversation, not a chatbot read aloud.
Read the guide →Everything else I work in
- Embeddings / Vector DBs
- Vector Stores
- Hybrid Search & Reranking
- Chunking Strategies
- Context Engineering
- Context Window
- Multi-Agent Systems
- Tool Calling
- Function Calling
- AI Orchestration
- Agent Memory
- Human-in-the-Loop
- LLM Gateways / Routing
- Guardrails
- Prompt Injection
- AI Hallucination
- Red Teaming
- Structured Output
- LLM Observability
- AI Governance
- Fine-tuning
- Prompt Caching
- Cost Optimization
- Latency Engineering
- Quantization
- Open-Weight Models
- Small Language Models
- Streaming Inference
- Voice AI Engineering
AI systems people can trust - safe by design, not by accident.
"Responsible AI" is usually a slide from a compliance team that never shipped a model. For me it's an engineering loop that runs on every release.
Every release, not once at launch
The engineering underneath.
An AI feature is only as good as the system it ships inside. This is the path a request actually takes - the screen it starts on, the server that takes it, the layers that answer it, and the data and infrastructure holding it all up.
- FrontendReactNext.jsAstroTypeScriptTailwind CSSshadcn/ui
- MobileReact NativeExpogluestack-ui
- BackendNode.jsExpressNestJSHono
- OrchestrationLangChainLangGraphLangSmithVercel AI SDK
- GatewaysBedrockOpenRouter
- ModelsOpenAIAnthropicGeminiGrokLlama
- RetrievalpgvectorPineconeVoyage AIBM25
- DatabasePostgreSQLMySQLSQLiteMongoDBRedisDynamoDB
- InfraAWSSupabaseVercelCloudflareDockerTerraform
Have an AI system that needs to actually work?
I take on a limited number of AI engineering engagements - RAG, agents, evals, or voice AI - each quarter. If you're evaluating whether your AI idea can survive production, let's talk.
Talk to Me About AI →