Agentic AI · MCP Integrations · Native iPhone Apps
Designed, built, and run in production.
Agentic AI · MCP Integrations · Native iPhone Apps
Designed, built, and run in production.
lexg.ai is my independent engineering practice. I build AI systems that connect language models to live data and real tools, and then I run them: agentic chat with per-request tool routing, MCP servers in production, inference that moves between a local GPU and cloud APIs, and the native iPhone apps that put it all in someone's pocket.
I work end to end — architecture, code, deployment, and operations — with the habits I picked up over 14 years keeping enterprise video and ad-tech systems honest: contracts before code, failure modes named before they happen, and nothing called done until it has run in production. I take on full builds and consulting engagements.
Local GPU inference with cloud failover. Models run on hardware you own when they can and fall back to OpenAI or Anthropic when they should, behind one interface
Learn more →
Model Context Protocol servers that give any MCP client scoped, authenticated access to your databases, files, and services — with tools planned per request, not dumped into every prompt
Learn more →
Python and Flask applications with AI at the core — agentic chat, voice dictation, document pipelines, and internal tools built for the people who'll use them every day
Learn more →
SwiftUI apps for AI-backed services — push notifications you can act on from the lock screen, typed API contracts, Keychain auth, and offline-tolerant requests
Learn more →
Coding agents on real codebases — parallel Claude Code sessions and multi-machine builds coordinated through committed contracts, with humans reviewing what ships
Learn more →
AI agents at every stage of delivery — specs, code generation, contract tests, CI/CD, and deployment — so teams ship faster while people stay in charge of the decisions
Learn more →
Seven years in server-side ad insertion at FreeWheel, now paired with AI — incident diagnostics, inventory reporting automation, and agents that read ad-server logs so engineers don't have to
Learn more →Swift • SwiftUI • APNs • Flask • OpenAPI Contract
The booking system my own service business runs on. A client requests a slot, my iPhone rings, and I approve it from the lock screen — the confirmation email is out before the phone is back in my pocket. Native SwiftUI app, Flask API, and a written contract both sides test against.
View case study →Python • Flask • MCP • OpenAI API • LM Studio • Whisper
One chat that reaches everything I run. A fast planner model picks the tool groups each request needs, then the chosen model works only with those MCP servers, on a local GPU or a cloud API, with Whisper dictation built in.
View case study →Python • LM Studio • NVIDIA GPU • OpenAI API • Anthropic API
Dual-backend inference that routes between a local NVIDIA RTX 5060 Ti and cloud APIs over a private WireGuard tunnel. Development and everyday workloads run on owned hardware; cloud models step in when the task or the load calls for them.
View case study →Python • MCP SDK • Streamable HTTP
A fleet of production Model Context Protocol servers for databases, files, web content, and vision. Any MCP client can use them, and a planner picks the tool groups each request actually needs instead of loading everything into every prompt.
View case study →Python • Vision-Language Models • Local-First
A local-first semantic image search system using vision-language models. Automatically describes, tags, and indexes images for natural language querying—no cloud dependencies required.
View case study →Python • C to Python • RAG • AST
RAG-powered code translation converting legacy C codebases to idiomatic Python. Dependency-aware context retrieval keeps translations semantically accurate, and local inference keeps the cost flat no matter how much code goes through it.
View case study →Python • NLP • Multilingual AI
Multilingual content generation on local LLMs via LM Studio — personalized, context-aware profiles produced in bulk without per-request API costs.
View case study →Python • ETL • Data Analytics
ETL pipelines that process UAD appraisal datasets into context-aware market analysis, turning raw property data into intelligence an investor can act on.
View case study →Knowledge Map • 2026 Edition
A linked reference map of the agentic AI ecosystem — foundation models, orchestration frameworks, coding agents, enterprise platforms, and the infrastructure that runs them in production.
Explore the landscape →Founder & Principal Engineer
I spent 14 years in production systems at Comcast, Charter, and FreeWheel, where I worked on the streaming ad pipelines millions of people watch through. In 2025 I went independent to build AI systems full-time.
Everything on this site is something I've built and run myself — including a native iPhone app I learned Swift to ship, which my own booking business now runs on. The longer version of my story is at lexgaines.com.
Have a system that needs to actually run? I take on end-to-end builds — AI platforms, MCP integrations, and native iPhone apps — as well as consulting engagements. Let's talk.