AI Customer Support Ticket Agent using Haystack

AI models and external services

• Qwen: Local LLM for chat responses (Qwen/Qwen2.5-1.5B-Instruct or 0.5B-Instruct)
• AI4Bharat IndicTrans2: Indian language translation (22+ languages)
• AI4Bharat Indic-Parler-TTS: Indian language text-to-speech
• Sentence Transformers: Knowledge base semantic search

Qwen, IndicTrans2, Indic-Parler-TTS, and sentence-transformers can run locally. Depending on configuration and feature use, the application also connects to RunPod/Sarvam, the Acrobuild CS API, SMTP, Edge TTS, Hugging Face model downloads, and administrator-supplied HTTP(S) knowledge sources.

Setup

pip install -r requirements.txt

Configuration

Create/update .env file with:

# Optional: HuggingFace token for AI4Bharat models (get from https://huggingface.co/settings/tokens)
HF_TOKEN=your_token_here

# AI4Bharat TTS Configuration
INDIC_TTS_DEVICE=cpu  # or 'cuda' if GPU available

Local-only chat does not require a commercial LLM API key. External integrations require their corresponding credentials and network access.

Run

uvicorn app:api --reload

The API will be available at http://127.0.0.1:8000

React Frontend

The React admin workspace lives in src/ and runs from the repo root.

npm install
npm run dev

With the backend running on http://127.0.0.1:8000, the React app will be available through Vite on http://127.0.0.1:5173.