| .claude-code-history | ||
| .forgejo/workflows | ||
| .sentry-native | ||
| .telemetry | ||
| androidApp | ||
| DATASETS/2026-04-09-BUSINESS-PANEL-SAMPLE-CONVERSATION/PiecesOS-LTM-Reports-SQlite-primary-dbus-hx-available | ||
| DOCS | ||
| gradle | ||
| litert-lm | ||
| scripts | ||
| shared | ||
| .env.example | ||
| .gitignore | ||
| build.gradle.kts | ||
| CHECKLIST.2026-05-06_pre-swarm-idempotency-refactor.md | ||
| CHECKLIST.md | ||
| CLAUDE.md | ||
| gradle.properties | ||
| gradlew | ||
| gradlew.bat | ||
| LICENSE | ||
| LOGS-DeepScriber-LiteRT-LM-experimental-builds-setup-and-runners-documentation-global-rules-refinements.txt | ||
| PROJECT_INDEX.json | ||
| PROJECT_INDEX.md | ||
| README.md | ||
| screenlog.0 | ||
| settings.gradle.kts | ||
| version.json | ||
| version.txt | ||
DeepScriber
The meeting assistant that knows what you need before you ask.
Proprietary and confidential. Copyright (c) 2026 Robin Cheung. All rights reserved.
Overview
DeepScriber is a privacy-first, on-device meeting assistant that runs on Android, Windows, and Linux. It captures audio, transcribes in real time, identifies speakers, and — most importantly — proactively surfaces relevant information during the meeting without being prompted.
Unlike competitors that require you to type a question, DeepScriber watches the conversation and anticipates what you need: citing case law during a legal discussion, flagging drug interactions during a medical consult, or surfacing client history during a sales call.
Key Features
- Anticipative Intelligence — AI surfaces relevant info proactively, no prompting needed
- On-Device Processing — STT, LLM, diarization all run locally; no cloud required
- Speaker Diarization — Know who said what (LS-EEND neural diarization)
- Knowledge Base — Ingest your documents (.md, .txt, .json); AI cross-references them live
- Meeting Notes — Auto-generated summaries, action items, decisions, open questions
- Cross-Platform — Single Kotlin codebase → Android + Windows + Linux
Tech Stack
| Layer | Technology |
|---|---|
| Language | Kotlin Multiplatform |
| UI | Compose Multiplatform |
| LLM | LiteRT-LM (Gemma-3n, Qwen3, Phi-4) |
| STT + Diarization | sherpa-onnx (Whisper, Parakeet, LS-EEND) |
| Entitlements | RevenueCat |
| Inference | ONNX Runtime (NNAPI / DirectML / CPU) |
Documentation
| Document | Purpose |
|---|---|
CHECKLIST.md |
Operational task list — the only doc a coding agent needs |
DOCS/ARCHITECTURE.md |
Phase I full spec: ERD, entities, services, UI screens |
DOCS/planning-roadmap.md |
4-phase roadmap with legal/medical/CRM connectors |
DOCS/competitive-analysis-hedy.md |
Hedy AI feature comparison |
DOCS/porting-analysis.md |
Platform feasibility matrix |
DOCS/module-mapping.md |
Module-by-module Swift → Kotlin mapping |
DOCS/architecture-diagrams.md |
10 Mermaid architecture diagrams |
DOCS/cleanroom-architecture-spec.md |
Full architecture specification |
Development Status
Phase I — Android In-Person MVP (in development)
License
Copyright (c) 2026 Robin Cheung. All rights reserved. See LICENSE.