Microsoft Developer’s Cozy Kitchen episode features Chef Govind Kamtamneni explaining how to build, customize, and fine-tune AI-driven software engineering agents using Azure AI Foundry, SpecKit, and practical math-first principles.

From Building to Fine-Tuning: Coding Agents that Optimize AI Workflows

Speakers:

  • Chef Govind Kamtamneni, AI & Software Engineering Expert (LinkedIn)
  • John Maeda, Host of Cozy AI Kitchen (LinkedIn)

Overview

This Cozy Kitchen episode explores the creation and orchestration of intelligent software engineering agents using Azure AI Foundry, GitHub, SpecKit, and advanced machine learning techniques like reinforcement learning. Chef Govind shares practical workflows and best practices, making it relevant for developers interested in both the fundamentals and advanced aspects of AI-native development.

What You’ll Learn

  • Customizable AI Project Templates: Bootstrapping projects using Azure AI Foundry templates.
  • Asynchronous Engineering Agents: How to design and orchestrate agents for complex, long-running software tasks.
  • Self-Hosting & SpecKit: Advantages of self-hosting AI Foundry and integrating with SpecKit for modular development.
  • Reinforcement Learning & Fine-Tuning: Approaches for refining agent behaviors and model performance using Hugging Face datasets and preference optimization.
  • Multi-Agent Orchestration: Strategies for parallelizing tasks and managing workflows across multiple agents.
  • Math-First Principles: The importance of mathematical foundations and academic research in AI development.

Chapters & Key Highlights

  • 00:01 – Starting with AI Templates
  • 00:40 – Customizing Templates in AI Foundry
  • 01:57 – Self-Hosting AI Foundry & SpecKit Features
  • 03:06 – Building Multi-Agent Software Workflows
  • 04:45 – Persistent Agents with GitHub Integration
  • 06:03 – Parallel Task Assignment for Agents
  • 08:15 – Model Fine-Tuning & Reinforcement Learning
  • 10:09 – Using Hugging Face Datasets
  • 13:00 – Essential Advice: Study Math & Read Papers

Tools, Platforms & Resources

Best Practices Shared

  • Emphasize modular templates for scalability
  • Choose self-hosted AI for greater control
  • Integrate tightly with GitHub and SpecKit for collaborative software workflows
  • Harness reinforcement learning and preference optimization for adaptive models
  • Track experiments and continuously test using open datasets
  • Build AI engineering skills by deepening knowledge of mathematics and research literature

Target Audience

  • Developers and engineers looking to adopt AI-native practices
  • Teams seeking real-world agent orchestration solutions on Microsoft Azure
  • Practitioners interested in ML/AI model customization and infrastructure

Further Exploration

  • Try AI Foundry and agent templates for rapid prototyping
  • Explore full Cozy AI Kitchen episodes for more hands-on demos and developer insights