Structured engineering evaluation of Pixeltable as the data and knowledge layer for a small RAG-style knowledge system.
python3 -m venv .venv
.venv/bin/python -m pip install -e .[dev]
.venv/bin/python scripts/create_sample_corpus.py
.venv/bin/python -m app reset
.venv/bin/python -m app ingest sample_data
.venv/bin/python -m app search "Which source discusses battery powered transport?"
.venv/bin/python -m app ask "How is battery powered transport supported?"| Area | Implementation target |
|---|---|
| Persistence | Pixeltable tables and local Pixeltable home |
| Document processing | Pixeltable pxt.Document and document_splitter |
| Chunking | Pixeltable view over source documents |
| Embeddings | Pixeltable embedding index with a module UDF |
| Retrieval | Pixeltable .similarity() search |
| RAG | Retrieval plus deterministic local answer synthesis, with optional provider extension |
| Provenance | Source document, chunk, retrieval, and answer records |
| Incrementality | Insert/update behavior observed through Pixeltable tables and indexes |