VocalOps

An AI-powered voice & CLI interface for inventory (DB) management.

I have built this project mainly to learn about DevOps & Agentic AI.

---

Stack

• Frontend: ReactJS, Vite, Tailwind CSS
• Backend & Database: FastAPI, SQLite
• AIML: Deepgram, LangGraph, LangChain, Groq, Ollama, ChromaDB, MLflow
• DevOps: DVC, Docker, Kubernetes, Prometheus, Grafana

---

Working

Audio is recorded in the browser, sent to Deepgram for transcription (Speech to text), and the transcript is routed through a LangGraph agent backed by two LLMs:

• Groq (llama-3.3-70b-versatile) - routes intent, generates read-only SQL and analysis, synthesizes responses.
• Ollama (vocalops-sql) - Finetuned using unsloth and colab. Generates mutating SQL (UPDATE, INSERT, DELETE) locally. Sensitive operations performed locally.

The agent uses six tools: check_order, update_order, create_order, delete_order, analyze_orders, and search_policies (RAG over a chromaDB vector store for queries on policies).

Browser mic → React UI ──┐
                          ├─ FastAPI → Deepgram STT ─┐
Terminal mic → Voice CLI ─┘                          │
                                             LangGraph Agent → Groq (router)
                                                  ├─ mutations → Ollama (local) → SQLite
                                                  ├─ analytics → Groq → SQLite
                                                  └─ policies  → ChromaDB (nomic-embed-text)

---

Running the App

Option 1: Web Interface

Start the FastAPI backend: python src/vocalops_voice/main.py

Start the ReactJS frontend: cd vocalops-frontend

npm install

npm run dev

Option 2: Voice/Text CLI (Terminal)

This mode includes Human-In-The-Loop (HITL) approval for all database changes. python src/voice_cli.py

• Type your query directly for text.
• Type v and Enter for voice recording.

---

Usage examples:

---

Setup

Prerequisites: Python 3.10+, Node 18+, Ollama running locally.

Clone the repository and setup the virtual environment

python -m venv venv

source venv/bin/activate

Install dependencies

pip install -r requirements.txt

Configure your environment variables

cp .env.example .env # DEEPGRAM_API_KEY, GROQ_API_KEY, OLLAMA_BASE_URL

Pull model weights tracked by DVC

dvc pull

Build the custom text-to-SQL model

cd data && ollama create vocalops-sql -f Modelfile && cd ..

Pull the embeddings model for vector search

ollama pull nomic-embed-text

Database initialization

python setup_db.py # creates and seeds data/orders.db

python src/vocalops_core/ingest.py # embeds policies into chromaDB

---