Roadmap

Where I am, where I'm heading

Latest update
6 Jun 2026Sanskrit / Gītā project kicked off. Corpus verified (701 verses · 16-voice commentary Council). Scope locked — grounded LLM+RAG, no model-training. Architecture drafted: wisdom-node schema · convergence law · citation-grounded generation. Next: Chapter 2 → nodes.

The streams

SANJAYA — Sanskrit / Bhagavad-Gītā
Today

Canonical Gītā corpus pulled + verified — 701 verses · IAST · word-by-word meaning · 5 English + 2 Hindi translations + a 16-voice commentary Council. Scope + architecture locked: grounded LLM+RAG, fidelity-checked, never invented.

Next
  • Chapter 2 → structured wisdom nodes
  • Purport (the WHY) synthesized across the 16 commentaries
  • Common-man simplification + Indian-language fan-out
  • Recitation (TTS) + a seeker's Q&A in their own language
VANI — Voice
Today

Push-to-talk dictation tool built + verified (faster-whisper, on-device). Vocabulary correction for domain terms.

Next
  • Live dictation in daily use (offline)
  • Read-aloud (TTS)
  • Speaker + prosody awareness (SHRUTI)
  • Long-form relay + on-demand ensemble
LIPI — Text
Today

Phase-1 pipeline built: detect → extract → classify (6 domains) → writer ID, across 22 Indian languages.

Next
  • Real-document integration testing
  • Wire as 2nd perception path
  • PaddleOCR (cheaper image OCR)
  • Authorship + tone analysis
Computer Vision · DRISHTI
Today

Architecture proven via an image-grading POC: grade, detect, confidence, orchestration, compliance gate.

Next
  • Image-grading to <5% human correction
  • MVP #1 Retail Out-of-Stock
  • MVP #2 Vehicle Damage
  • AXIOM governance MVP

The SANJAYA pipeline

A śloka, carried to the common person in every Indian language — fidelity-checked at every step, never invented. Read through three lenses:

M1 Segment per-verse records supplied
M3 Verify checked vs a canonical source build
M4 Transliterate Devanāgarī → IAST supplied
M5 Anvaya word-by-word meaning supplied
M6 Literal plain Sanskrit → English / Hindi supplied
M7 Purport the WHY, across 16 commentaries build
M8 Fan-out into every Indian language build
M9 Simplify for the common person build
M10 Recite read-aloud (TTS) rent
M11 Nodes structured wisdom nodes build

build = our work · supplied = already in the public-domain corpus · rent = an existing tool. Our build is just the four "build" steps — everything else is given or bought.

Horizon

Now → 1 monthVoice dictation in daily use · LIPI integration testing · Gītā Chapter 2 into wisdom nodes · website as the live shopfront
1 → 3 monthsFirst productized CV vertical (Retail OOS) · TTS + SHRUTI senses · Gītā simplified into 2–3 Indian languages · Personal AI Twin intent capture
3 → 6 monthsSecond CV vertical + AXIOM · cross-modal intelligence · all 701 verses, multi-language + recitation · multi-user readiness
⌂Home
🔊Om
🎙Ask ✦Vision ◷Roadmap