# Turning every customer call into explainable coaching data.

> A real-time sales call evaluation system that converts streaming speech into structured, evidence-backed quality scores.

Author: [Dharani Eswaramurthi](https://dharani-ai-portfolio.vercel.app/)
Canonical page: https://dharani-ai-portfolio.vercel.app/work/voice-intelligence
Category: Speech AI · Sales intelligence
Outcome: 30% manual → 100% automated coverage

## The challenge

Manual review covered only a fraction of calls and produced inconsistent feedback. The system needed to listen in real time, preserve the supporting evidence, and make every score auditable.

## Engineering choices

1. Built a streaming speech pipeline with Soniox STT and AWS SageMaker for low-latency processing.
2. Structured the evaluation around observable call evidence instead of opaque, unsupported scores.
3. Designed the output for both fast operational scanning and deeper coaching review.

## The result

Audit coverage moved from 30% manual review to 100% automated evaluation, giving sales teams consistent, evidence-backed feedback across the full call set.

## Technology

Python, Soniox STT, AWS SageMaker, FastAPI, WebSockets, Structured outputs

## About the evidence

This is the author's project account. The reported outcomes are portfolio claims, not an independent benchmark. Interface illustrations and architecture walkthroughs are explanatory visuals, not live customer data.
