Samarthya CRM — AI-Powered Voice & Lead Platform

Outbound sales and citizen outreach teams frequently experience severe productivity loss due to manual dialing delays, inconsistent conversation note-taking, and subjective lead scoring. Teams require an automated voice outreach pipeline that transcribes calls in real time and objectively calculates prospect sentiment to prioritize follow-up.

Full-stack Developer and AI Architect. Built the backend telephony integrations, real-time transcription and analysis pipeline, and responsive React frontend dashboard.

A modern web application built on Node.js and React. Telephony calls are initiated concurrently through Twilio Voice. Audio streams are piped to Deepgram for low-latency live transcription. The text transcript is fed into a LangChain pipeline backed by Hugging Face NLP models to extract sentiment metrics, detect objection points, and recommend next actions.

Implemented scalable webhook handlers managing parallel outbound voice calls with fine-grained status tracking. Integrated Deepgram's streaming speech-to-text API for immediate transcript generation. Formulated structured NLP prompts and sentiment classifiers to evaluate customer interest levels during conversations. Built an intuitive dashboard in React and Tailwind CSS providing visual call timelines, sentiment breakdown meters, and filtered lead management.

Recognized as a Top Finalist among 505 competing teams at the BPUT Hackathon 2025. Successfully demonstrated real-time speech transcription and automated sentiment categorization under live demonstration conditions.