Bolorey (Vaak)

AI-driven Corporate Communication Coach

Bolorey helps professionals speak with clarity, confidence, and authority in corporate settings. Unlike generic public speaking apps, Bolorey focuses on real workplace scenarios — project updates, client calls, stakeholder meetings, and executive briefings.

The system is designed to be drillable (Board → Phases → Categories → Tasks), mirroring how corporate training is rolled out: start lean with an MVP, then add feedback loops, roleplay, analytics, and enterprise integrations.

---

🌟 Vision

• Private practice sandbox — zero fear of judgment, unlimited reps.
• Real-time nudges — filler detection, pacing, assertiveness feedback during sessions.
• Post-session learning — regression over time, personalized weak-spot tracking.
• Enterprise-ready — team dashboards, SSO, HR/LMS integration.

Bolorey turns your communication growth into a structured roadmap: 📋 Board (Vaak) → 🎯 Phases → 📂 Categories → ✅ Tasks

---

🏗️ Architecture Overview

Phase 1 — Practice Mode (Backend Core)

Recorder (sounddevice)
   ↓ chunks
Whisper ASR (faster-whisper)
   ↓ transcript
Analyzers (fillers, pace, pauses, tone)
   ↓ analysis
Feedback layer (metrics + session summary)
   ↓
Regression (store transcript + metrics)

Phase 2 — Interactive Trainer

Recorder → ASR (low-latency Whisper)
   ↓ (parallel)
Analyzers (fast checks) → Feedback Nudges
   ↓
LLM Trainer (persona: manager/client) ↔ User
   ↓
Session Summary (analyzers + LLM reflection)
   ↓
Regression (store transcript + metrics)

---

🚀 Roadmap (Phases)

1. MVP (Phase 1)

   • Audio capture & storage
   • Whisper ASR transcription (streaming, low-latency sliding window)
   • Rule-based analysis (fillers, pacing, pauses, tone)
   • REST API endpoints (session lifecycle, chunk ingest, live partial, stats, finalize)
   • SQLite (or DuckDB) storage

2. Interactive Trainer (Phase 2)

   • Roleplay with simulated managers/clients (served by future orchestration layer)
   • Low-latency ASR + feedback nudges (via analyzers + streaming endpoints)
   • Privacy controls (API + data governance)

3. Regression Learning (Phase 3)

   • Track progress over time
   • Dashboards with trendlines
   • Adaptive nudges

4. Portability & Future-Proofing (Phase 4)

   • Export/import transcripts (JSON/Parquet)
   • Semantic search over past sessions
   • Archival storage + compliance

5. Advanced Extensions (Phase 5)

   • Multi-language support
   • Real meeting integration (Zoom/Meet plugins)
   • Personalized growth plans
   • Advanced analytics

6. Deployment & Rollout (Phase 6)

   • CI/CD pipeline
   • Enterprise integrations (SSO, LMS, compliance)
   • Pilot programs
   • Pricing & commercialization

---

📦 Dependencies

Runtime (requirements.txt)

• faster-whisper — ASR model
• ffmpeg-python — audio I/O (requires ffmpeg via brew)
• sounddevice, soundfile — recording + WAV handling
• fastapi, uvicorn, pydantic, python-multipart — API layer
• sqlalchemy, pandas, duckdb, pyarrow — storage + analytics
• numpy, scipy — math, resampling
• python-dotenv, tqdm, typing_extensions — utilities

Dev/Test (requirements-dev.txt)

• pytest, pytest-cov, pytest-watch
• black, isort, flake8
• mypy
• pre-commit (optional)

Lock file

• requirements.lock.txt (generated via pip freeze) → ensures reproducibility in CI/prod.

---

🧪 Testing & TDD

• Tests live in tests/ (pytest).
• TDD approach: stubs created for analyzers, feedback, regression → failing until implemented.
• Test data: tests/data/hello_16k.wav.
• Watch mode for fast dev:

  make tdd
  

• Coverage:

  make coverage-html
  open htmlcov/index.html
  

---

🏰 Moats (Differentiation)

• Analytics: filler detection, pacing, regression dashboards.
• UX Drills: interactive trainer, adaptive exercises (implemented in separate UI repo).
• Coaching Logic: structured corporate framing, growth plans.
• Platform Edge: optional infra (low-latency, semantic search, compliance).
• Human Network: not in scope (BetterUp’s domain).

---

🛠️ Developer Workflow (Backend)

• Setup:

  python3 -m venv .venv
  source .venv/bin/activate
  pip install -r requirements.txt -r requirements-dev.txt
  

Makefile commands:

make test           # run tests
make tdd            # run tests in watch mode
make coverage       # coverage in console
make coverage-html  # coverage in browser
make lint           # style + static checks
make format         # auto-format (if configured)
make clean          # remove caches

Run the API server

python -m vaak.api.server

Example endpoints:

curl -X POST localhost:8000/asr/session
curl -X GET  localhost:8000/health

The web UI now lives in the separate vaak-ui repository (see that repo's README for frontend setup, dev commands, and deployment details).