Syllabus
Six courses, one project. Each course adds new capabilities to the same system — mirroring how real-world AI applications grow over time.
LDB-001 · LLM Fundamentals in Python
Kickstart your repo and ship a minimal tech-support assistant using pure Python and the OpenAI SDK (no third-party libs). Understand tokenization, context windows, and how cost/latency affect quality.
You’ll build: a CLI/web script that answers support questions and preserves short conversational context.
LDB-002 · Prompt Engineering
Design prompts that clarify, summarize, and adapt to user preferences. Add lightweight telemetry to track token usage and spot regressions.
You’ll build: reusable system/user prompt templates + a small “prompt budget” dashboard.
LDB-003 · Retrieval-Augmented Generation (RAG)
Ground answers in your own documents and fail gracefully with fallbacks and “I don’t know (but here’s what to try)” patterns.
You’ll build: a simple RAG pipeline (ingest → chunk → embed → retrieve) and an eval script to compare answer quality.
LDB-004 · Guardrails for LLM Apps
Align behavior with business rules, safety, and compliance. Filter risky queries, enforce tone/policy, and log moderation events.
You’ll build: a guardrail layer (policy checks + moderation hooks) that runs before/after model calls.
LDB-005 · From Assistant to Agent
Go beyond chat: add actions and escalation. Trigger webhooks, post to Slack, and hand off to humans when confidence is low — prioritizing critical cases.
You’ll build: Slack alerts + webhook actions with a scoring system that routes issues faster.
LDB-006 · Tool Use with Reasoning Loops & MCP
Implement plan-act-observe loops so your assistant can call tools, analyze results, and iterate toward a solution. Use Model Context Protocol (MCP) to register tools via a common interface.
You’ll build: a reasoning loop that chains multiple tool calls and an MCP tool registry your agent can query at runtime.
By the end, you’ll have
- A working agent repo (prompts, RAG, guardrails, tools, evals).
- Ops basics: cost/latency tracking, safety logs, and smoke tests.
- Integrations: Slack + webhook-based actions you can extend.
- A shareable demo and README for your portfolio.
Ready to start building?
One project. Real feedback. Your pace.