Chitthi: A Séance Where the Ghost Can Only Quote Itself

Description and use

Chitthi lets users ask historical figures questions, answering only with verbatim passages from scanned letters, ensuring authenticity and transparency while running entirely offline.

Problem

Historical figure chatbots often generate fabricated quotes, undermining authenticity. Users need a system that answers only with verbatim passages from original letters.

Outcome

Chitthi delivers authentic, verifiable answers from scanned letters, avoiding fabricated content. It runs entirely offline on a laptop, uses open models, and provides citations and original scans for transparency.

Workflow

• Scan PDF/image/text → render pages with pypdfium2 or Gemma 4 vision → extract passages (≤900 chars, page‑bound) → embed passages with EmbeddingGemma and index in SQLite FTS5 → receive user question → semantic search + exact‑word search → Gemma 4 generates answer in archivist or séance mode → checker verifies quoted spans against passages → UI displays answer, highlights verified quotes, flags missing quotes, links to original scan

Evidence

Chitthi reads scanned letters page by page, indexes them on your own machine, and answers only from the passages it retrieves

Source

Every quoted phrase is checked word-for-word against the letters. Verified quotes are highlighted; a quote that isn't in the letters is flagged.

Source

Everything in the demo is the real app running against Gemma 4 on a 16 GB M4 Mac. The only edit is cutting the 40 to 60 seconds Gemma takes to think.

Source