Chunking, hybrid search, structured outputs, tool calling. The week where the pieces from Week 1 become a system that answers questions about your own documents.

Five build days, two live sessions, one ship day, one read day. Monday, October 5 to Sunday, October 11.

What this week is actually for

Week 1 gave you components. Week 2 assembles them into RAG, then immediately shows you why the basic version of RAG fails in production and what to do about it.

Most tutorials stop at Day 9, the naive pipeline. That pipeline breaks the first time a user searches for a part number. Days 10 through 12 are the difference between a demo and something a client will pay for.

Day Date Focus Type
8 Mon, Oct 5 ‣ Build
9 Tue, Oct 6 ‣ Build
10 Wed, Oct 7 ‣ Build plus Live Session 3
11 Thu, Oct 8 ‣ Build
12 Fri, Oct 9 ‣ Build plus Live Session 4
13 Sat, Oct 10 ‣ Ship Day
14 Sun, Oct 11 ‣ Read Day

Live sessions this week

Session 3, Wednesday October 7, 7:30 PM IST, 90 minutes. Vector search fails live on a part number, then we fix it with Reciprocal Rank Fusion.

Session 4, Friday October 9, 7:30 PM IST, 90 minutes. We give a model two real Python tools and build the execution loop that every agent framework sits on.

Tech Support runs every day at 9:00 PM IST on Discord.

What you have at the end of the week

Production RAG 2.0: chunking with overlap, hybrid retrieval with RRF, Pydantic-validated structured extraction, and a working tool dispatcher. Plus a benchmark table with real latency numbers from your own machine.

That is Portfolio Build 2 of four, and it is the one that gets you interviews.

The one thing that trips people up

Day 12 is the hinge of the entire thirty days. Tool calling is what separates a chatbot from an agent, and everything in Week 3 assumes you have it working.

If Day 12 does not land, do not push forward into Week 3. Bring it to the Friday session or the Saturday Tech Support call and get it working first.