Tom di Mino
Applied AI/ML Engineer
Full-stack AI engineer with 4+ years shipping LLM-powered production systems. Built $800k/day AI tool at JP Morgan serving 5,000+ wealth advisors. Currently architecting production AI advisor platform at SubQ AI with fine-tuned LLMs, voice pipelines, and multi-dimensional evaluation frameworks. Startup builder who moves fast, empathizes, and ideates constantly.
Experience
Subquadratic AI
Principal AI/ML Engineer
Startup building AI advisors that internalize expert methodologies via fine-tuned and post-Transformer LLMs
- Architected production SMS/mobile AI platform with 9 cognitive agent "souls" and layered memory
- Deployed fine-tuned LLMs to production via TensorRT-LLM + Triton on 4×H100 with tensor parallelism
- Built LLM-as-a-Judge evaluation framework using Gemini, Qwen3, Kimi K2 for 9-dimensional persona assessment
- Implemented voice-first pipeline with sub-second latency, barge-in detection, and speculative RAG
- Designed cognitive flows for AI advisors that remember across sessions and emotionally attune
- Maintained internal Claude Code skills repo and mentored teams on agentic planning, debugging, and development
JP Morgan Chase
AI Content Engineer
World's largest bank, building AI tools for 5,000+ wealth advisors
- Built "Meeting Assist" AI tool generating $800k/day savings with 72% adoption rate
- Designed 5-agent agentic architecture using LangGraph, evaluated with DeepEval
- Reduced hallucinations in production by rewriting prompts and deconstructing chain-of-thought
- Developed RAG pipelines using LlamaIndex and proprietary solutions
Sr. Content Designer / AI UX Copywriter
Worked with AI UX, RAI, and DeepMind teams
- Designed curriculum for Googlers advancing technical leadership skills (Markdown, HTML/CSS)
- Interviewed AI/UX leaders and synthesized learnings into internal/external thought leadership
- Audited mentorship platform serving 22k+ employees, drove 25% rise in program launches
Chan Zuckerberg Initiative
Sr. Content Designer
Building open-source scientific tools for image analysis
- Co-led design of learning platform for napari—Python-based image analysis for biologists
- Published full course on GitHub using Jupyter Book, Jupyter Notebooks (Markdown, Python, and narrated video)
- Collaborated with 9 application scientists, UX researchers, and SWEs to ship MVP in 6 months
Notable Projects
Claude Code Starter-Kit
Open-source ~/.claude/ configuration—40 skills, 30+ slash commands, 7 lifecycle hooks, CLI tools, and a VS Code extension. Covers local ML, RAG, browser automation, academic research, agentic teams, and session tracking.
View on GitHub →
Claudicle
Open-source alternative to OpenClaw with four-layer soul architecture (soul.md, Cognition, Memory, Channels)—configurable personas, cognitive pipelines, and Slack integration. Built for augmenting Claude Code sessions.
View on GitHub →
BG3 Script Extender macOS
Open source native ARM64 port of Baldur's Gate 3 Script Extender. Ground-up systems programming—runtime hooking, memory injection, embedded Lua scripting. 94% feature parity with 1,999 components via Ghidra reverse engineering.
View on GitHub →
Kothar wa Khasis
AI oracle living inside minoanmystery.org—perceives visitor behavior, remembers across sessions, generates tarot cards, and speaks through RAG-enhanced memory. Built on the Open Souls paradigm. Named for the Canaanite craftsman god.
Speak with Kothar →