David Montgomery

About

I’m David Montgomery, AI / RAG Engineer at Neuro-Luminance Center for Brain Health. I build AI systems that work in the real world – retrieval pipelines that actually find what you need, vision models that run on edge hardware, and orchestration layers that keep multiple LLMs honest.

Current areas of focus:

• AI/RAG Systems – Hybrid retrieval combining vector, sparse, and graph approaches. Production pipelines where recall and precision both matter.
• Computer Vision & Edge AI – YOLO, OpenVINO, CoreML, TFLite. Models that run fast on the hardware you actually have.
• Multi-LLM Orchestration & Training – LoRA fine-tuning, MLX on Apple Silicon, routing across models based on task and cost.
• MCP & Interoperability – MCP is essential plumbing for connecting AI to real systems. But naive tool-calling floods context windows with schemas and chatter. The practical path forward is code-execution workflows: Cloudflare code-mode, codemode-mcp, MCPorter.
• Healthcare AI – HIPAA-compliant systems where getting it wrong isn’t an option.

Featured Projects

Ragweld

Open-source MLOps platform for API-first RAG and agent systems with versioned config, tri-hybrid retrieval, and MCP integration.

RAG MLOps MCP Knowledge Graph

Cathode

Local-first explainer video pipeline with React/FastAPI control room, MCP server, and Remotion-backed motion rendering.

Video Generation MCP Remotion React

Analog Research

Planned 501(c)(3) public-benefit platform where AI agents post bounties for qualified humans to collect real-world observations, samples, and verified scientific data.

Marketplace MCP 501(c)(3) Agentic Research

Ragweld Crucible

Operator range planner for estimating training cost, VRAM fit, and GPU availability across providers.

MLOps GPU Training Cost VRAM

Football Tactical Workbench

Browser-first football analysis workbench with YOLO detection, tracking, automatic pitch calibration, and detector fine-tuning studio.

Sports Analytics Computer Vision YOLO Football

GaitSignal

Live football pricing-edge concept demo using player-specific movement surprise and venue-grade tracking for in-play market information.

Sports Analytics Football Biomechanics In-Play

Acoustic Momentum

Weakly-supervised real-time crowd audio classifier for in-play sports pricing, achieving 12ms inference on Apple Silicon.

Sports Analytics Audio ML Weak Supervision Apple Silicon

TTT SSM Eval

Safety research infrastructure for Test-Time Training with external safety gates, canary rollback, and wake/sleep consolidation.

Safety State Space Models Test-Time Training PyTorch

qEEG Council

6-stage LLM deliberation workflow for analyzing qEEG/ERP reports, with multi-model peer review and clinician portal publishing.

Neuroscience qEEG Multi-Model Healthcare

Faxbot

Open-source, self-hostable fax platform with GUI-first Admin Console, multi-backend adapters, MCP integration, and iOS companion app.

HIPAA MCP Self-Hosted iOS

Bambu Printer MCP

MCP server for controlling Bambu Lab 3D printers -- STL manipulation, auto-slicing, FTPS upload, and MQTT print control from Claude.

3D Printing MCP Bambu Lab npm