SelectedWorkSelected Work

A production AI calling platform for Payal Gold. Voice agents answer inbound calls 24/7 and run outbound campaigns, then send each customer the right catalogue on WhatsApp. The sales team runs everything from one real-time console.

1. Distributed outbound dialer on a Postgres job queue (SKIP LOCKED), so parallel workers never double-dial a customer.
2. Self-learning call concurrency (AIMD) that discovers the carrier's hidden line limit on its own.
3. LLM post-call pipeline that extracts outcome, interest and objections, then sends matching products on WhatsApp via pgvector search.

A multi-tenant SaaS that reads handwritten driver sign-in sheets, reconciles them against ADP timecards, flags every mismatch, and follows up with drivers through AI voice calls and messages. Operators approve before anything reaches a driver.

1. Two-stage OCR (layout detection, then LLM adjudication) that turns handwritten sheets into structured shifts.
2. The AI writes, deterministic code decides: plain-English rules compile once into validated conditions.
3. Durable Temporal workflows for driver follow-up, with retries and manager escalation.

Patients call the clinic and speak to Vitara, who identifies them, verifies their identity, registers new patients and books, reschedules or cancels appointments. Emergencies and difficult calls go to staff.

1. A 20-node LangGraph state machine: code controls the conversation, the LLM only does narrow jobs at temperature 0.
2. Multi-step identity check: caller ID, then date of birth, then name or phone fallback.
3. OpenAI-compatible streaming endpoint so the agent plugs straight into Vapi as a custom LLM.