Notes on LLM memory, RAG, Darija NLP, and building products in Morocco.
2026-09-12
8 min
Flat memory can't answer relational questions like 'how has my relationship with X been trending?'. The entity-layer design behind iRemember: alias resolution across languages, evolving profiles, sentiment timelines, and privacy by construction.
2026-09-05
Naive summarization destroys exactly the details users ask about later. The compression strategy behind iRemember: episode-vs-fact split, progressive oldest-first gating, temporal anchor preservation, and a probe-question evaluation method.
2026-08-29
9 min
A teardown of the extraction pipeline behind iRemember: why LangGraph, how a chat message becomes structured memory candidates, deduplication, async processing, and the schema decisions that make retrieval work.
2026-08-22
11 min
Everything required to build production memory for LLM agents: the four memory operations, hybrid retrieval scoring, write control, history compression, person-entity modeling, and how 2026 benchmarks actually compare.
2026-07-12
6 min
Context windows keep growing, but they still forget. Why persistent, structured memory — not longer contexts — is what makes LLM assistants actually useful.
2026-05-03
5 min
Tenant isolation sounds like a database decision. It's actually an application-wide discipline. Lessons from shipping a multitenant ERP.
2026-02-18
7 min
Darija breaks every assumption standard NLP pipelines make: no standardized orthography, code-switching mid-sentence, and almost no labeled data. Field notes from building sentiment and transcription tools.