Intelligence

What Lala knows — memory store, retrieval stats, and indexing health

Total chunks

684

Source files

7

Students indexed

7

Messages covered

4,598

Chunks added · last 30 days

0

New embeddings ingested into the RAG corpus per day.

Coverage leaders

Students with the most indexed context

Recently ingested

Spot-check the newest chunks added to the corpus

dataset/distilled/selin.txt · chunk #5 · 8 messages

dataset/distilled/selin.txt · chunk #8 · 6 messages

dataset/distilled/selin.txt · chunk #2 · 5 messages

dataset/distilled/selin.txt · chunk #4 · 4 messages

dataset/distilled/selin.txt · chunk #6 · 8 messages

dataset/distilled/selin.txt · chunk #7 · 8 messages

dataset/distilled/selin.txt · chunk #0 · 5 messages

dataset/distilled/selin.txt · chunk #1 · 8 messages

dataset/distilled/selin.txt · chunk #3 · 6 messages

dataset/distilled/selin.txt · chunk #9 · 6 messages

How Lala uses this

Every Lala message runs a semantic search over this corpus with match_coaching_chunks (pgvector HNSW, cosine similarity). The top matches are injected into the prompt so Lala can reference past coaching patterns when advising the student.

684 chunks stored as vector(768) embeddings.
Sourced from 7 files covering 7 unique students.
Run npm run ingest to refresh.