Labs
Core labs (1–8) are the main graded path. Labs 9+ are optional extensions that deepen specific 2026 topics (structured output, caching, GraphRAG, guardrails, etc.). Complete them in order where possible — many build on earlier labs.
ChatBot with FastAPI
Build a streaming chat API with FastAPI, Dockerize it, deploy to Render, add Google Auth for protected endpoints, embed a simple frontend on HuggingFace Spaces.
RAG ChatBot
Build a full RAG pipeline: upload PDFs, chunk, embed into Chroma, retrieve with hybrid search, re-rank, generate answer, evaluate with RAGAS.
Gemma4 Finetuning
Finetune Gemma4 on a custom instruction dataset using QLoRA with Unsloth, push adapter to HuggingFace Hub, create a Spaces demo.
Hybrid RAG ChatBot
Extend Lab 2 with PGVector, add metadata filtering, multi-query retrieval, and a Streamlit/Gradio UI.
Signature Detection & Cropper
Use Grounding DINO Tiny to detect and crop signature regions from scanned documents, output cropped PNGs + bounding box JSON.
AI Agent + MCP
Build an autonomous research agent: given a question, it searches the web, reads URLs, stores findings in SQLite, and synthesizes a report via MCP server.
CI/CD Pipeline
Add a full GitHub Actions pipeline (lint → test → build Docker → push to GHCR → deploy to Cloud Run) with branch protection and NeMo Guardrails.
Full MLOps on GCP
End-to-end: Cloud Function trigger → GCS ingest → BigQuery ML → Cloud Run serving → Cloud Logging monitoring → automated retraining.
LLM Red-Teaming
Build a repeatable red-team harness for a tool-using LLM app and turn exploits into CI regression tests.
Instructor Extraction Loops
Build a schema-first extraction pipeline with retries, validation, and tests.
Prompt Caching & Cost Benchmarks
Implement a safe prompt/response cache and quantify cost + latency wins.
Contextual Retrieval Upgrade
Make chunks self-contained with headings/metadata and measure RAGAS improvements.
GraphRAG Mini
Build a tiny knowledge graph alongside embeddings and use it for multi-hop retrieval.
LangGraph Agent Workflow
Implement a stateful agent graph with retries, guard conditions, and logs.
Polars + DuckDB Local Warehouse
Build a fast local analytics pipeline (ETL → Parquet → SQL features) for ML/RAG.
Guardrails + Red-Team Regression
Add guardrails to an LLM app and prove they work with a red-team suite in CI.