AI-Powered Calling Agent & Hyperlocal Analytics Platform

VoxReach is a full-stack AI calling agent platform that lets businesses run human-like voice conversations at scale — placing outbound cold calls, handling inbound support queries, and making reservations — while surfacing hyperlocal analytics on call volume, resolution rates, and regional trends to guide business decisions.

Built the calling infrastructure on Twilio for telephony, with Cartesia and Sarvam handling low-latency text-to-speech and speech-to-text so the agent could hold natural, human-like conversations, including support for Indian languages. Designed the call flow to handle interruptions, silences, and turn-taking to keep interactions feeling conversational rather than scripted.

Implemented a retrieval-augmented generation pipeline so the agent could pull accurate, business-specific answers (pricing, policies, availability) in real time instead of relying on static prompts. Routed inference across Groq, Gemini, and open-source models based on latency and cost trade-offs, keeping response times low enough for live phone conversations.

Built the backend on FastAPI with MySQL and TiDB for storing call transcripts, outcomes, and structured metadata at scale. Built a Node.js service layer for orchestration and a Next.js dashboard for area-wise call analytics — visualizing volume, resolution rates, and regional trends — with the full stack deployed on AWS for reliability and scale.

Automated end-to-end voice interactions that previously required human agents, cutting response time for customer queries and reservations from minutes to seconds. Hyperlocal, area-wise call analytics gave the business visibility into demand patterns across regions, directly informing outreach and resourcing decisions. The RAG-based knowledge layer kept agent responses accurate and current without manual script updates, and the multi-LLM routing (Groq/Gemini/open-source models) kept inference costs low while preserving response quality across call volumes.