Study embeddings, semantic similarity, chunking, and vector…

1. 01

   Concepts & system design

   60 min· read · map the mechanismtimer

   R11 — Hugging Face embedding model cards & revisions

   R13 — Qdrant collections & payload metadata

   Read the assigned sections; map the mechanism, threat model and trade-offs.

2. 02

   Project implementation

   120 min· build · integrate · debugtimer

   Study embeddings, cosine/dot-product similarity, chunking strategies, and vector indexes. Build P2's local ingestion prototype (50–100 pages of sample docs).

   Build, integrate, debug and improve the increment in the active project.

3. 03

   Security lab

   60 min· attack your own systemtimer

   Inventory embedding model, tokenizer files, config, Python packages, and artifact sources; generate initial SBOM manifest (D12).

   Test the abuse case, failure path, authorization bypass or resource limit.

4. 04

   Verification & evidence

   30 min· test · record evidencetimer

   Working embedding pipeline + initial D12 artifact and dependency manifest.

   Add or run regression tests; record the 9-point evidence record.

5. 05

   Deliberate practice & Socratic review

   30 min· explain aloud · update gapstimer

   Answer the 8 Socratic questions without notes: What? Why? How? What if? Why not? How can I prove it? Where can it break? What have I learned?

   Re-explain the hardest concept aloud, inspect one edge case, update the gap log.

6. total5 h

• R11Hugging Face Documentation↗ huggingface.co

  Model cards, tokenizer/config files, embedding usage, and repository commit-hash revision pinning.

• R13Qdrant Documentation↗ qdrant.tech

  Collections, payload metadata indexing, pre-filtering, multitenancy partitioning, and snapshots.

Read only the named sections needed for today's task. Reading sits inside the 60-minute concepts block; it is not extra time.

• docsQdrant Documentation · reference docs
  Indexing - Qdrant

  Explains why payload indexes must exist before ingestion and how filter-aware HNSW edges change query planning, which is the mechanism behind metadata-filtered vector search you build in the ingestion prototype.

  R13moved 2026-11-11
• videoVenelin Valkov · 18 min
  Why Your RAG Gives Wrong Answers (And 4 Chunking Strategies to Fix It) | LangChain Text Splitters

  Walks through structure-aware, semantic, LLM-driven and contextual-enrichment chunking with LangChain splitters, mapping directly onto the chunk-boundary decisions in the 50-100 page ingestion prototype.

  R11checked 2026-11-11
• videoSatyajit Das · 8 min
  Vector Search and HNSW Explained Visually

  Visual walkthrough of the layered HNSW graph and the ef/m/efConstruction knobs, so you can explain why approximate nearest-neighbour recall changes as your index grows.

  R13checked 2026-11-11
• articleHugging Face Blog · tutorial article
  Getting Started With Embeddings

  Shows the concrete mechanics of encoding text with a sentence-transformers model, comparing vectors with cosine similarity versus dot product, and ranking results by semantic distance.

  R11checked 2026-11-11
• repoQdrant (GitHub) · notebook collection
  qdrant/examples: A collection of examples and tutorials for Qdrant

  Runnable notebooks including qdrant_101_text_data that ingest real documents end-to-end, giving you a working skeleton for P2's local ingestion prototype.

  R13checked 2026-11-11

A lab is complete only when it produces evidence. Installing PyRIT, Garak, Presidio or TransformerLens is not completion. Saved on this device only.