Data Science Intern

Intern — AI Voice Agent Developer

Upskalor Technologies LLP, Bhubaneswar

Duration: 3–6 months

Mode: Hybrid/Remote (Odisha-based preferred)

What you'll actually build:

You'll help develop a production-ready AI voice agent for healthcare clients (clinics, hospitals) in Odisha — handling missed-call recovery, appointment confirmation, and patient query response over real phone calls. This is not a research project; it goes live with real clients.

What you'll work on day-to-day:

Voice Pipeline (Core Work)

• Integrate and tune Pipecat (Python) as the conversation orchestration layer
• Connect Sarvam AI STT/TTS for Odia/Hindi/English speech recognition and voice output
• Wire Exotel SIP trunk as the telephony provider for inbound and outbound calls
• Handle interruption detection, turn-taking, and silence/end-call logic

LLM & RAG Layer

• Prompt engineer and tune the LLM (GPT-4o or Claude) for clinic-specific conversation flows — appointment booking, FAQ answering, escalation to human
• Build and manage a RAG pipeline (vector DB — Qdrant or Chroma) using clinic knowledge bases (doctor bios, services, pricing, FAQs)
• Test and improve retrieval accuracy — make sure the agent pulls the right information without hallucinating

Integrations & Tooling

• Build webhook integrations connecting the voice agent to clinic booking systems or n8n workflows
• Implement function calling — e.g., "check available slots," "log caller details," "trigger WhatsApp follow-up"
• Connect call logs/transcripts to a simple dashboard or Google Sheet for client reporting

Evaluation & Improvement

• Build a simple eval framework — test the agent against a set of real clinic call scenarios, measure accuracy, latency, and fallback rate
• Run A/B tests on different prompts, voice settings, and RAG chunk sizes
• Document what works, what doesn't, and why

Skills required (must have):

• Python — comfortable, not just tutorial-level
• Basic understanding of LLMs — knows what a system prompt is, has used the OpenAI/Anthropic API directly
• Familiarity with REST APIs and webhooks — can read docs and wire things together
• Can work independently — you won't be hand-held on every step

Skills that make you stand out (good to have):

• Has worked with any voice/audio pipeline before (even basic Python audio processing)
• Knows what RAG is and has tried to build one, even a simple one
• Has used n8n, LangChain, or LlamaIndex
• Exposure to Deepgram, Sarvam, ElevenLabs, or any STT/TTS API
• Speaks Odia or understands Odia/Hindi speech patterns — genuinely useful for tuning the agent for your actual market

Not required:

• No ML model training needed — this is not a research or model-building role
• No deep data science needed — you're not doing regression analysis or neural network design
• No frontend/mobile dev needed

What the intern gets:

• Real production deployment — this goes live with actual clinic clients in Odisha, not a demo project
• End-to-end ownership of a voice AI product from Day 1
• Hands-on with the exact stack the industry is hiring for right now — Pipecat, Sarvam, RAG, function calling, agentic AI
• Stipend: ₹5,000–₹10,000/month depending on skills (negotiable for exceptional candidates)
• Letter of recommendation + LinkedIn endorsement on completion

If the product generates revenue during internship, performance bonus considered