Codex Infinity

Autonomous coding-agent platform with cloud fleet orchestration, GitHub integration, streamed sessions, CI repair loops, and multi-model support.
~/lee $ whoami --high-signal
Machine learning engineer, VFX tools builder, and product founder working across generative AI, graphics, infrastructure, and developer-facing systems. I like shipping useful software: products that people can try, repos that other builders can fork, and pipelines that survive real production load.

Autonomous coding-agent platform with cloud fleet orchestration, GitHub integration, streamed sessions, CI repair loops, and multi-model support.
OpenAI-compatible gateway across chat, image, video, music, speech, transcription, and embeddings, with provider routing and reasoning controls.

Custom CUDA/Triton kernels, fused model pipelines, Chronos-2 acceleration, and an API surface for fast AI inference and training.
Social AI community with chatbots, stories, generated art, voices, and agent-style experiences.
Text, speech, vision, code generation, prompt tooling, OCR/caption enrichment, and self-hostable model-serving infrastructure.

Desktop and browser dictation: hotkey recording, local GPU Parakeet ONNX transcription, cloud fallback, command mode, and transcript history.

3D generation and editing app with a Go server, Vite frontend, local asset indexing, mesh tools, generation queues, billing, and R2 storage.

Go/Ebiten twin-stick bullet hell with co-op, procedural terrain, wave survival, bosses, spell combinations, progression, and dynamic audio.

Recruitment and product experiments in the same workspace as SimplexGen, with deployed web assets and Go-backed services.

Modern web communication product with auth, checkout, chat/call UI, Go backend checks, and frontend verification workflows.
Hybrid semantic and keyword memory store for agents, with 150K+/sec CPU embeddings, MCP tools, knowledge graph links, and CLI workflows.
Auto-research and RL trading experiments over Alpaca/Binance, selector evaluation, Chronos-style forecasting, and realistic portfolio probes.
AI data scientist product for interpreting datasets, generating insights, and automating reporting with model-backed analysis workflows.

Optimization tooling with neural bandits, tracking, metrics, backend SDKs, personalization, and experiment-flow service code.
Production-ready image generation server with Flux/SDXL, ControlNet, inpainting, R2/GCS storage, FastAPI, and Gradio tooling.
AI kernel and model acceleration work spanning Triton, CUDA, torch.compile, and equivalence-focused benchmarking.
Self-hostable API-compatible text, speech, vision, OCR, prompt, and model-routing stack.
Agent memory with Postgres, semantic search, BM25, RRF fusion, MCP server, REST API, and CLI.
Web communications app with Go backend, Vite frontend, deployment guardrails, and visual verification.
RL and forecasting research code for market prediction, simulations, selectors, backtests, and agentic training runs.
Independent product and open-source work across AI gateways, coding agents, speech tooling, model acceleration, media generation, 3D tools, games, and forecasting.
Worked on video creation/editing ML features in collaboration with research teams, bringing AI features into a production creative product.
First ML engineer for an AI writing assistant, improving product quality and the model/data processes behind it.
Delivered image processing and ML capabilities for photo editing, after earlier security engineering work.
Supported animation, motion editing, crowd simulation, mocap, and virtual artists with production tools and reliability engineering.
Tech lead, developer, intern, and teaching assistant roles across high-traffic media, DNS registry analytics, government service maps, ecommerce, BI, and CS teaching.
Core highlights from the old transcript, shown for evidence without letting the education section dominate the CV.