RAG Systems
Production retrieval built for real users
Grounded assistants with retrieval, source-aware answers, and architectures designed for reliability instead of demos.
Generative AI Engineer
Building production-ready GenAI systems with strong engineering foundations.
I work across RAG, tool-connected assistants, evaluation, contextual memory, and observability to build AI products that are useful, measurable, and reliable in real environments.
AI Twin
Ask about my work
A conversational layer for discussing architecture, evaluation, systems, and project experience.
Suggested prompts
How was the RAG app evaluated?
What guardrails did you add?
How do you approach production AI?
About
Applied AI engineering with full-stack depth.
My background combines production software engineering with recent deep work in Generative AI. I have built systems across agritech, geospatial, construction, and AI product workflows, with a focus on turning technical capability into practical, measurable delivery.
Evaluation
Quality loops that make AI accountable
Offline datasets, baseline scoring, regression checks, and practical feedback loops for improving output quality over time.
Observability
Visibility into what the system is doing
Tracing, structured logs, metrics, and operational thinking that make debugging and iteration faster in production.
Experience
A path from shipped product work to applied GenAI systems.
2026 - Present
Independent Generative AI Engineer
Building production-style GenAI systems across RAG, evaluation, guardrails, contextual memory, and observability.
2024 - 2025
AI First Software Developer
Shipped AI-powered geospatial product features for agricultural monitoring workflows used across multiple farms and clients.
2022 - 2024
Software Engineer / Full-Stack Developer
Delivered data-heavy interfaces, upload workflows, and cloud-connected spatial products across construction and real estate.
Signals
Evidence over claims.
0.78 faithfulness baseline established through RAG evaluation
50% longer supported conversations via context management and memory strategies
45% faster map rendering by moving to COG tile workflows
Full-stack delivery across AI, geospatial, cloud, and product-facing systems
Projects
Case studies from real systems.
These projects are presented as engineering case studies rather than generic portfolio cards. The focus is on architecture, evaluation, observability, product usefulness, and what was actually built.
FastAPI, Pinecone, OpenAI, Ragas, AWS
Production RAG System
A production-oriented retrieval system with offline evaluation, retrieval-vs-generation benchmarking, Prometheus and Grafana observability, and grounded answer generation.
0.78 faithfulness baseline0.62 answer relevancy baseline0.75 context recall baseline
Built to show what a production-minded RAG stack looks like beyond a demo, with measurable quality and operational visibility.
LangGraph, FastAPI, Postgres, Pinecone, Phoenix
Agentic Legal Review Backend
A production-style multi-agent backend for legal review and workflow automation, combining retrieval, orchestration, persistence, human approval, and offline evaluation.
Multi-agent review orchestrationPostgres-backed run persistenceHuman-in-the-loop revision flow
Built to demonstrate a more realistic agentic backend with traceability, workflow state, review lifecycle persistence, and evaluation support.
Portfolio Expansion
More Applied AI Systems
This portfolio structure is set up to expand with more GenAI, evaluation, and internal tooling work as additional systems are published.
More live systems to addMore architecture breakdownsMore project write-ups coming
The goal is to turn this site into a stronger technical portfolio rather than a generic personal landing page.
Blogs
Writing on practical GenAI engineering.
This area can evolve into articles, technical notes, or architecture breakdowns. I have added professional placeholders so the structure is already there as you publish more work.
Building RAG Beyond the Demo Stage
Thoughts on grounding, evaluation, and why a working prototype is not the same thing as a production-ready assistant.
Drafting / Coming soonContextual Memory for Long-Running AI Interactions
How memory, trimming, and compaction patterns change the usability of assistants over longer conversations.
Drafting / Coming soonWhat I Measure in Applied GenAI Systems
A practical view on evals, observability, and the signals that matter when you want AI systems to be trusted.
Drafting / Coming soon