Context-first agentic systems
Isaac Flath
I help teams build retrieval-first AI products: context pipelines, evals, traces, and workflows for agents that need reliable private knowledge.
Jake Selvey
Director of Analytics & Data Science
Centriam
“If I had to pick one individual to assist me in tackling an unknown data-related problem, large or small, it would be Isaac.”
What I Do
I help teams turn AI ideas into agents that find the right context, act reliably, and fit inside real work.
Useful agents need three layers working together: context that matches the task, control that keeps actions reliable, and workflows that people can actually adopt.
What the agent knows, retrieves, remembers, and sees at the moment it acts.
How tools, constraints, evals, and feedback loops keep the system reliable.
Where the agent becomes useful: inside the products and processes people already use.
Builder
Systems work in practice
I work through the tools I write about: retrieval systems, agent workflows, publishing pipelines, and product experiments.
Teacher
Courses on practical AI
My courses turn technical systems into repeatable workflows, from RAG fundamentals to AI-assisted development.
Writer
Years of public thinking
The archive shows years of adjacent work across ML, deep learning, AI writing, web development, and product experiments.
For teams
I am available for retrieval-first AI product work.
I help teams building AI products over private knowledge inspect and improve the context pipeline: retrieval, memory, evals, traces, citations, and workflow fit.
Audrey Roy Greenfeld
Co-founder, Feldroy
Co-creator of Cookiecutter; co-author of Two Scoops of Django
“I'm using AI to build everything I've dreamed of building, and it's in big part thanks to these teachings.”
Context, Control, Workflows
One system for agents that find the right information, act reliably, and live inside real products and processes.
Context
Help agents find the right information, because that solves most of the problem.
Control
Give agents the tools, constraints, evals, and handoffs they need to act reliably.
Hugo Bowne-Anderson
Data & AI scientist, writer, educator, podcaster
“I'll read anything Isaac writes: it's high signal, makes me think, and helps me understand what's happening in AI.”
Teaching and Courses
These courses show how I turn messy AI systems into workflows people can understand, practice, and use.
Active course
Learn RAG
A practical course on retrieval augmented generation, from keyword search and vector search to multimodal retrieval, chat interfaces, and citation generation.
Prior course
AI-Assisted Coding
A course on AI-assisted development. The materials and testimonials show the same pattern: systems, constraints, feedback loops, and practical workflow design.
Emily Ekdahl
Founding AI Engineer
Stealth Startup
“I became radically more effective in how I build with AI agents.”
How I Got Here
My background combines workflow reality from operations, technical depth from software and data, and clarity from teaching.
Operations and Process
Learned to see work as systems: handoffs, bottlenecks, incentives, and the gap between a clean plan and what happens on the floor.
Data, Software, and ML
Built the technical depth to connect data, software, models, and product constraints without losing sight of how people will actually use the system.
Teaching and Agent Workflows
Teaching forced me to make complex tools explainable. That now shapes how I build agent systems: context, control, and workflows that teams can understand, adopt, and trust.
A Personal Note
I met my wife, Alyssa, when we were both teaching dance full time. Dance has been one of the great joys in my life, and it is still a shared part of our story.
She now runs Art of Movement, a dance instruction business in the D.C. area.
Jelle de Jong
Data Specialist
Waterschap Drents Overijsselse Delta
“I can now add features more easily, hunt down and fix bugs faster, and make my code more secure and reliable.”
Weekly article
Get the weekly deep-dive on context, control, and workflows for useful agents.
5,000+ readers