8 micro-topics in AI Systems, each with the evidence that proves you have it and the written reason behind every link. Before you start, it rests on 5 other domains. Downstream, it holds up 1 domain.
8 topics · depth 12–14 of 16 · 8 internal links · 7 in · 2 out
Load-bearing links first. Each one says what in this domain rests on what outside it, and why.
2 links
Measure retrieval quality rests on Build an evaluation dataset
Metrics are computed over an evaluation set.
Measure generation faithfulness rests on Use a model as a judge
Faithfulness is scored by a judge.
2 links
Design a retriever-generator system rests on Store and query vectors in Postgres
Retrieval needs stored vectors.
Combine keyword and semantic search rests on Filter retrieval by metadata
Fusing results requires filtered vector search.
1 link
Design a retriever-generator system rests on Treat the context window as a budget
Retrieval is a decision about what goes in the window.
1 link
Inject retrieved context rests on Maintain chat context
Retrieved text is placed into the request.
1 link
Combine keyword and semantic search rests on Query rows with SELECT
The keyword half is a text query.
In prerequisite order. Each idea names the artifact that closes it, the market demand that put it on the map, and what it stands on.
Depth 12 · conceptual · unassisted · M19 §19.1
Separates retrieval from generation
Store and query vectors in Postgres · Embeddings & Vectors
Retrieval needs stored vectors.
Treat the context window as a budget · Context Engineering
Retrieval is a decision about what goes in the window.
Depth 13 · procedural · unassisted · M19 §19.1
Places retrieved text where the model will use it
Design a retriever-generator system
Injection is the generation half of the design.
Maintain chat context · Model Integration
Retrieved text is placed into the request.
Depth 13 · conceptual · scaffolded · M19 §19.3
Runs hybrid retrieval and fuses results
Query rows with SELECT · SQL
The keyword half is a text query.
Filter retrieval by metadata · Embeddings & Vectors
Fusing results requires filtered vector search.
Retrieval quality work starts once the pipeline runs.
Depth 13 · conceptual · guided · M19 §19.3
Improves recall for badly phrased questions
Design a retriever-generator system
Query rewriting sits in front of retrieval.
Depth 13 · meta · scaffolded · M19 §19.4
Applies precision at k, recall at k and MRR
Design a retriever-generator system
You measure a system that exists.
Build an evaluation dataset · LLM Evals
Metrics are computed over an evaluation set.
Depth 14 · conceptual · guided · M19 §19.3
Refines the top-k with a stronger model
Combine keyword and semantic search
Reranking refines a candidate set.
Depth 14 · representational · scaffolded · M19 §19.5
Attributes each claim to its source chunk
Citations point at the chunks you injected.
Depth 14 · meta · scaffolded · M19 §19.4
Detects answers unsupported by context
Use a model as a judge · LLM Evals
Faithfulness is scored by a judge.
Faithfulness compares an answer to its context.
Retrieval quality is measured before generation quality.
Everything downstream that names an idea here as a prerequisite, grouped by where it lives.
2 links
Defend against prompt injection rests on Inject retrieved context
Retrieved content is untrusted input in the prompt.
Prevent PII leakage rests on Inject retrieved context
Retrieval can pull personal data into a prompt.
Retool. Retrain. Relaunch.
375 ideas. 17 weeks. No tuition, ever.
Vets Who Code is a veteran-run 501(c)(3). The accelerator is free, remote, and we don’t take a share of your first paycheck.
Free · Remote · 17 weeks · EIN 86-2122804