Raman Shrivastava
← Teaching

Track 01

Build AI Agents

From transformer foundations to production agent operations.

8 modules · 10 lessons published · pinned a4648b1

Module 01Transformer FoundationsStrong basics are non-negotiable before building agents.Module 02RAG (Retrieval-Augmented Generation)The first enterprise use case of LLMs since 2022. Immediately useful at work.Module 03Agentic WorkflowsRigid workflows with some agentic control + MCP integration.Module 04soonFull Agents (Human-in-the-Loop)Agent has full control plus human oversight and hooks for safety.
  • Agent autonomy and the autonomy ladder
  • Human-in-the-loop interrupts and approvals
  • Safety hooks and policy enforcement
Module 05soonSkills and SandboxesThe next 3-6 months of agent evolution.
  • The Claude Skills pattern (capabilities packaged as files)
  • Sandboxed code execution for agents
  • When 'agent skills' is just rebranded prompt + tools
Module 06soonCoding AgentsAGENTS.md plus tools plus repo structure for AI-first development.
  • The CLAUDE.md / AGENTS.md pattern for instructing agents
  • Hooks, permissions, environment configuration
  • Iteration planning across long-running coding tasks
Module 07soonPersonal AgentsAlways-on agents that live in the messengers you already use.
  • Daemon agents vs per-invocation CLIs (OpenClaw vs Claude Code)
  • Telegram and WhatsApp gateway architecture
  • Kubernetes deployment of personal agents
Module 08soonAgent OperationsObservability, evals, security, cost, and reliability for agentic systems.
  • Tracing and observability: span design for agentic loops
  • Cost engineering: token accounting, prompt caching ROI
  • Production evals and failure forensics

All the code, runnable demos, and the live agent stay in build-ai-agents. MIT-licensed, open, regenerated when the repo changes.