LLM API craft
Chat and completion APIs, system prompts, structured JSON output, tool calling, streaming and token budgeting.
The technical builder track for developers and SDETs who want production-style LLM systems — retrieval pipelines, agents over MCP, evaluation harnesses and parameter-efficient fine-tuning — not a Playwright testing syllabus.
Eight focus areas. Each ends with a working notebook or repo you can keep.
Chat and completion APIs, system prompts, structured JSON output, tool calling, streaming and token budgeting.
Few-shot patterns, chain-of-thought where it helps, guardrails, and failure modes when prompts drift.
Embedding models, chunk strategies, overlap, metadata filters and hybrid lexical + vector search.
Ingest → index → retrieve → rerank → generate with citations; hallucination checks before answers ship.
ChromaDB and comparable stores: collections, indexing choices, freshness and cost at modest scale.
Multi-step agents, MCP server design, tool allowlists and human checkpoints for risky actions.
DeepEval, Ragas and LangSmith-style suites: golden sets, regression gates and red-team probes.
When to fine-tune vs prompt vs RAG; dataset prep and LoRA / PEFT trade-offs.
Builders work through in-house projects like these.

Resume analysis and improvement suggestions powered by LLM feedback with structured critique outputs.

Ask questions over large PDFs with a retrieval-augmented study assistant and source citations.

Batch sentiment analysis of customer reviews for product teams, with theme clustering.

Meeting assistant that summarises discussions and extracts action items with owners.
Guided labs in every class, not pre-recorded video.
Ganesh and Moksha teach from the systems they ship.
Each focus area ends in a repo you keep.
Fees and batch dates are shared on a quick call.
We will email batch dates and an invite to a demo session. Fees are shared on a call.