# Cole McIntosh

> Cole McIntosh is an AI engineer in Dallas, Texas and the founder of Mellow AI, building production AI agents, document extraction, and retrieval systems in Python and TypeScript.

AI engineer and founder of Mellow AI building production agent systems, document extraction pipelines, and developer tools in Python, TypeScript, and Rust. Combines hands-on systems work—from fine-tuned small models and zero-dependency parsers to multi-provider LLM infrastructure—with end-to-end delivery that measurably reduces operational load.

Roles: Founder at Mellow AI (2025–Present); AI & Data Engineer at Pay Ready (2023–2025); Scaled Tooling Analyst at Cloud Kitchens (2022–2023); Business Analyst at RealPage (2021–2022).

Open source: LiteLLM — Shipped SSO admin API and UI, Google Cloud Model Armor guardrails, Langfuse OpenTelemetry tracing, usage/health dashboards, provider integrations (Hyperbolic, Lambda AI), Gemini caching and Perplexity search cost accounting, and Ollama streaming fixes.; Pydantic AI — Implemented xAI search tools and attachment lifecycles, model-agnostic X search via subagent fallback, OpenAI input token counting, UI adapter URL sanitization, and provider host/timeout settings.; smolagents — Fixed Enum and custom metaclass execution in sandboxed agent code, added AWS Bedrock API-key support, and refactored MultiStepAgent test coverage..

Based in Dallas, Texas. Contact: cole@staymellow.ai. Site: https://colemcintosh.io. This file may be quoted; attribute to Cole McIntosh and cite the source URL.

## Pages

- [Home](https://colemcintosh.io/): Who he is, what he ships, projects, open source, and work history.
- [Blog](https://colemcintosh.io/blog): Essays on AI agents, retrieval, extraction, evals, and MCP.
- [Resume](https://colemcintosh.io/resume): Work history, projects, open source, and certifications.

## Resume in machine-readable form

- [Markdown](https://colemcintosh.io/resume.md): The full resume as text.
- [JSON](https://colemcintosh.io/resume.json): The same fields, structured.
- [PDF](https://colemcintosh.io/api/resume): The printable version.
- [MCP server](https://colemcintosh.io/mcp): Streamable HTTP, no auth. Point Cursor, Claude Code, or VS Code at this URL to read the resume as tools and resources.
- [RSS feed](https://colemcintosh.io/rss.xml): New essays.

## Writing

- [Everything Is Computer](https://colemcintosh.io/blog/everything-is-computer): Harbor turns jobs into containers, instructions, and graders. The bottleneck is the test. Published 2026-08-27. Topics: Harbor, AI agents, Terminal-Bench, Agent evaluation, Developer tools. Markdown: https://colemcintosh.io/blog/everything-is-computer.md
- [Every Max Coding Agent Plan, Tested](https://colemcintosh.io/blog/every-max-coding-agent-plan): Four $200 coding agents on the same work. Which surface makes it disappear? Published 2026-08-17. Topics: Coding agents, Claude Code, Codex, Cursor, Developer tools. Markdown: https://colemcintosh.io/blog/every-max-coding-agent-plan.md
- [Python to TypeScript in One Evening](https://colemcintosh.io/blog/porting-openextract): Parallel agents and fast inference made the port possible. Deciding what exists stayed hard. Published 2026-08-14. Topics: TypeScript, Python, openextract, AI SDK, Zod. Markdown: https://colemcintosh.io/blog/porting-openextract.md
- [Build AI That Removes Work, Not Adds It](https://colemcintosh.io/blog/ai-efficiency-not-more-work): Untrusted AI creates more work. Measure total efficiency, not capability. Published 2026-03-28. Topics: AI strategy, Automation, Evals, Product engineering. Markdown: https://colemcintosh.io/blog/ai-efficiency-not-more-work.md
- [When Code Is Easy, Judgment Is Scarce](https://colemcintosh.io/blog/code-and-judgment): When AI makes code cheap, context, review, and judgment become scarce. Published 2026-01-18. Topics: AI engineering, Code review, Judgment, Technical leadership. Markdown: https://colemcintosh.io/blog/code-and-judgment.md
- [Building langchain-salesforce](https://colemcintosh.io/blog/langchain-salesforce): How I built a LangChain integration for Salesforce CRM. Published 2025-11-09. Topics: LangChain, Salesforce, Python, Open source. Markdown: https://colemcintosh.io/blog/langchain-salesforce.md

## Projects

- [Openextract](https://github.com/Mellow-Artificial-Intelligence/openextract): I built a library that turns documents, images, audio, and video into validated Pydantic objects, across model providers.
- [pdfmd](https://github.com/colesmcintosh/pdfmd): I built a zero-dependency PDF-to-Markdown converter, including the PDF parser, DEFLATE decoder, and layout reconstruction.
- [Liedar](https://github.com/Mellow-Artificial-Intelligence/liedar): I built a document screening tool that combines file metadata, copied-region detection, and model assessments to surface signs of tampering.
- [TeraCode](https://github.com/colesmcintosh/teracode): I built a local-first Rust TUI that turns a development goal into an isolated, multi-agent task DAG with explicit workspace and autonomy gates.
- [mev](https://huggingface.co/stay-mellow-ai/mev): I fine-tuned Qwen3.5-4B to choose one of 2–24 options from a structured state, and published the weights, training details, and evaluation.
- [LiteLLM](https://github.com/BerriAI/litellm/pulls?q=is%3Apr+is%3Amerged+author%3Acolesmcintosh): Contributed model and provider support, streaming fixes, and usage reporting to the LLM gateway.
- [Pydantic AI](https://github.com/pydantic/pydantic-ai/pulls?q=is%3Apr+is%3Amerged+author%3Acolesmcintosh): Added xAI search and attachment support, OpenAI input token counting, and provider configuration improvements.
- [smolagents](https://github.com/huggingface/smolagents/pulls?q=is%3Apr+is%3Amerged+author%3Acolesmcintosh): Fixed Enum and custom metaclass execution, added Bedrock API-key support, and improved agent structure and tests.

## Optional

- [llms-full.txt](https://colemcintosh.io/llms-full.txt): Every essay and the resume in full, in one file.
