🌌 CognifyAI — Advanced RAG + Speech AI Assistant

CognifyAI is an intelligent assistant that combines Retrieval-Augmented Generation (RAG) with Speech AI to provide fast, accurate insights from your documents and audio.
Built with Streamlit for a beautiful interactive UI and Supabase for persistent storage, CognifyAI is designed for document understanding, audio transcription, semantic search, and conversation memory.

Core idea: enable a lightweight web app that listens, reads, remembers, and answers — backed by vector retrieval and modern generative models.

---

📑 Table of Contents

• Highlights
• UI Screenshots
• Tech Stack
• Quick Start
• Supabase Schema
• Docker
• Performance
• Troubleshooting
• Security
• Roadmap
• Contributing
• Credits

---

🚀 Highlights

• 🎙️ Real-time audio recording (browser) + transcription
• 🔍 Hybrid search — semantic + keyword retrieval for precise answers
• 🧾 Document ingestion with chunking and embeddings (RAG)
• 🗣️ Multi-speaker diarization support (optional)
• 🧭 Contextual conversation memory persisted in Supabase
• 🕒 Timeline & entity extraction and exportable summaries
• 🐳 Deployable via Docker, Streamlit Cloud, Heroku, GCP, or AWS

---

📸 UI Screenshots

Documents Upload	Recordings & Recorder	Quick Tools	Supabase Console

---

🧩 Tech Stack

• Frontend / App: Streamlit (Python)
• Storage / DB: Supabase (Postgres + Storage)
• Models / APIs: Google Gemini (optional), Whisper (OpenAI or local), HuggingFace Transformers, Cohere (optional)

---

📦 Repository Layout

/
├─ assets/                # screenshots
├─ requirements.txt
├─ hybrid_search.py
├─ rag_app.py
├─ speech_processor_enhanced.py
└─ README.md              

---

🛠️ Quick Start (Local)

1. Clone

git clone https://github.com/VikasKumar281/CognifyAI.git
cd CognifyAI

2. Install dependencies

python -m venv venv
source venv/bin/activate      # Windows: venv\Scripts\activate
pip install --upgrade pip
pip install -r requirements.txt

3. Create environment file

cp .env.example .env
# Edit .env and add API keys and Supabase credentials

Important environment variables

# Required / important
GOOGLE_API_KEY=
SUPABASE_URL=
SUPABASE_KEY=

# Optional
HUGGINGFACE_TOKEN=
COHERE_API_KEY=
WHISPER_MODEL=base
STREAMLIT_SERVER_PORT=8501

4. Run

streamlit run rag_app.py
# Open http://localhost:8501

---

🗄️ Supabase Database Schema

Use the following SQL to create the required tables in Supabase (Public schema):

-- Conversations table
create table if not exists public.conversations (
  id bigserial primary key,
  session_id text not null,
  query text not null,
  response text not null,
  context text,
  timestamp timestamptz default now()
);

-- Document metadata table
create table if not exists public.document_metadata (
  id bigserial primary key,
  filepath text not null,
  document_type text,
  chunk_count integer,
  processed_at timestamptz default now()
);

-- Give your anon/service roles permission
grant usage on schema public to anon, service_role;
grant all on public.conversations to anon, service_role;
grant all on public.document_metadata to anon, service_role;

⚠️ For production, lock down permissions and use row-level security (RLS) and role-based access instead of granting all to anon or service_role.

---

🐳 Docker (Recommended for reproducible deploys)

Dockerfile (example) :

FROM python:3.9-slim
RUN apt-get update && apt-get install -y ffmpeg curl build-essential && rm -rf /var/lib/apt/lists/*
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY . .
RUN mkdir -p temp outputs
EXPOSE 8501
CMD ["streamlit", "run", "rag_app.py", "--server.port=8501", "--server.address=0.0.0.0"]

docker-compose.yml (example)

version: '3.8'
services:
  cognifyai:
    build: .
    ports:
      - "8501:8501"
    volumes:
      - ./temp:/app/temp
      - ./outputs:/app/outputs
      - ./.env:/app/.env
    restart: unless-stopped

Run:

docker-compose up --build -d

---

⚡ Performance & Best Practices

• Limit conversation context window (e.g., last 8–12 messages) to reduce token usage.
• Chunk large documents (CHUNK_SIZE ≈ 512–1000 tokens) and cap total chunks per doc.
• Cache expensive model loads (Streamlit @st.cache_data) and I/O operations.
• Use smaller Whisper model (tiny or base) for faster transcription on low-resource instances.
• Use connection pooling for database access (e.g., psycopg2.pool).

Example caching:

@st.cache_data
def load_whisper_model(model_name="base"):
    return whisper.load_model(model_name)

---

🐛 Troubleshooting

• Streamlit not starting — ensure port is free; try --server.port=8502.
• Audio issues — verify ffmpeg is installed and accessible.
• Supabase connection errors — check SUPABASE_URL and SUPABASE_KEY.
• Model OOM / memory errors — use smaller models or increase instance RAM.

---

🔒 Security & Production Notes

• Never commit .env or API keys to Git. Use platform secrets.
• For public deployments, enable HTTPS and configure CORS and XSRF protection.
• Implement authentication (Supabase Auth or OAuth) before exposing features that modify data.
• Add rate-limiting and monitoring for model/API usage.

---

♻️ Backup & Monitoring

• Schedule DB backups (pg_dump) and store artifacts in cloud storage.
• Add basic performance metrics to the sidebar (processing time, memory).
• Log user activity (with privacy in mind) for debugging.

---

🛣️ Roadmap & Ideas

• Add user authentication & RBAC (Supabase Auth)
• Integrate advanced RAG retrievers (FAISS/Chroma) for local vector storage
• Multi-language support for transcription and search
• A/B tests between different LLM providers for cost/perf tradeoffs

---

🙌 Contributing

Welcome contributions. Please:

1. Fork the repo
2. Create a feature branch (feat/…)
3. Open a PR with a clear description & tests (if applicable)

---

📬 Credits & Contact

Developed by Vikas Kumar — (GitHub: VikasKumar281)
If you want me to also create release-ready assets (banner, compressed screenshots, or a GH Actions workflow), I can add them to the repo.

Enjoy building — CognifyAI brings memory to assistants. 🌌