Louis-François Bouchard is the co-founder and chief technology officer of Towards AI and the creator of What’s AI, an educational platform for people building practical AI systems. He focuses on the engineering decisions that determine whether language-model applications remain accurate, affordable, and useful beyond an initial demonstration.
Bouchard developed his career in Montreal, studying engineering at École de technologie supérieure and pursuing doctoral research in artificial intelligence and computer vision at Mila and Polytechnique Montréal. He also worked in applied AI and led AI development at designstripe.
He launched the YouTube channel behind What’s AI in 2020, initially explaining research papers before expanding into language models, retrieval, and AI engineering. He co-founded Towards AI in 2022 and left doctoral research in 2024 to concentrate on education and company-building. His open-source guides range from machine learning for beginners to practical AI engineering and large language models.
With fellow Towards AI co-founder Louie Peters, he co-authored Building LLMs for Production, a 2024 guide to production language-model applications covering prompting, retrieval, evaluation, fine-tuning, agents, deployment, and optimization.
Autonomy should match the problem. Bouchard distinguishes direct prompting, retrieval, predetermined workflows, autonomous agents, and multi-agent systems according to the control, cost, and flexibility each task requires. For one marketing application, he replaced a proposed multi-agent architecture with a single agent and specialized tools, preserving shared context and reducing handoff errors. His research-and-writing workflow similarly separates exploratory research from tightly constrained writing and human editorial judgment.
Context engineering must account for prompt caching. Working on Towards AI’s open-source tutor, Bouchard challenged the assumption that shrinking conversation history automatically saves money. Summarization can destroy cached context, discard important details, and trigger repeated retrieval; retaining the complete conversation sometimes improves cost, latency, and recall simultaneously. His context-engineering analysis emphasizes evaluating retention, retrieval, and compaction against actual application behavior.
AI Research OS makes research memory reusable. With Paul Iusztin, Bouchard built a file-based research-memory system that organizes notes, videos, repositories, and other materials into Markdown files, indexes, and lightweight research wikis. Agents can recover relevant knowledge across projects without requiring a vector database for every personal workflow. The presentation’s 10,994-note collection belongs to Iusztin, not Bouchard.
Through Towards AI’s courses, enterprise training, and AI engineering mentorship, Bouchard teaches engineers to measure reliability, preserve human oversight, and choose the simplest architecture that meets production requirements.
Combine exploratory research, file-based handoffs, structured editorial feedback, and a calibrated judge to turn a human brief into technical content you can inspect and improve.
Why a fluent post still needs research and editing
An AI tutor’s compaction experiments show how cached history, retrieval, and hardware limits change what belongs in the context window—and when removing it becomes necessary.
The instruction was there. Why did the agent ignore it?
A file-based research system connects saved notes to current work, then preserves sources, summaries, and new questions in a project wiki that can keep growing.
▶ Watch ↗
