CODLAB / Engineering
Engineering Notes.
How do you separate data, verify answers and check changes? Notes on the technical decisions behind Django applications and local AI systems.
Every article: a concrete problem, a checked example and the limits of the approach.
Browse the notesAll notes
Four new articles · three revised notes-
Django: bind data access to the tenant
A document ID is not a permission. Check membership, scope queries and test IDs belonging to other tenants.
Django · ORM · PostgreSQL -
RAG: sources as untrusted data
Local inference does not stop instructions inside documents. Bound the context, authorise sources and validate citations.
Python · RAG · On-premise -
Django i18n: check language, URLs and metadata
Translated text is not enough. Language switching, internal links and SEO metadata must describe the same page edition.
Django · i18n · SEO -
Playwright and axe: a check that can fail
Check accessibility on the rendered page. Scan desktop and mobile and prove that the check detects a failure.
Playwright · axe-core · CI -
Evaluate RAG: separate retrieval from answers
Recall, hit rate and answer correctness measure different things. A small example and a reproducible verification plan.
Python · Retrieval · Regression -
Llama on GPU: budget memory, then measure
Model weights are only part of memory use. Check KV cache, context and runtime before selecting a serving configuration.
Llama · GPU · vLLM -
pgvector with Django: check filters and recall
A tenant filter limits visible data. The execution plan determines whether ANN search finds enough results.
PostgreSQL · pgvector · Django