Signal AI

Turn Every Sales Call Into Revenue Signals

Evidence + Volume = Action

VWO AI Hackathon 2025

Sales Teams Are Drowning in
Unstructured Call Data

📞

15-20 Minutes

per call review

Manual analysis doesn't scale

🔍

Critical Bugs

buried in noise

100+ weekly transcripts to sift through

💸

Lost Revenue

from missed signals

Competitor threats & churn risks go unnoticed

⚠️This costs companies millions in missed opportunities

What If AI Could Read
Every Call in <1 Minute?

🐌
15 min

Manual Review

NEW
⚡
<1 min

Signal AI

0x

Faster Than Manual Analysis

Extract bugs, features, competitors & action items automatically

6 Specialized AI Agents
Working in Parallel

6
Agents
🐛
Bug Hunter
Catches critical issues
✨
Feature Finder
Spots customer needs
🎯
Competitor Intel
Tracks market threats
📊
Case Study Spotter
Finds success stories
📡
General Signals
Sentiment & urgency
📝
Call Summary
Participants & actions
⚡All agents run simultaneously for maximum speed

<1 Minute Analysis Pipeline

1
📝

Input

Paste transcript or upload audio

2
🎯

Triage

Embedding-based priority routing

3
🧠

Analysis

6 agents extract signals in parallel

4
📊

Scoring

ML model calculates priority

5
✅

Output

Actionable insights with evidence

<1 min
Total Processing Time
End-to-end analysis

Cutting-Edge AI Architecture

Real innovation, not just API wrappers

🎯

Intelligent Triage

Embedding-based routing skips LLM when possible

  • ✓Multi-signal similarity scoring
  • ✓70%+ accuracy without API calls
  • ✓Instant priority classification
🔑

Multi-Key Optimization

6 dedicated Gemini API keys for zero rate limiting

  • ✓One key per agent
  • ✓Parallel execution = 6x throughput
  • ✓Automatic fallback system
🧠

ML-Powered Scoring

GradientBoosting model, not hardcoded formulas

  • ✓ONNX export for production
  • ✓R² = 0.73 accuracy
  • ✓Trained on real call data
Next.js 16
Gemini 2.5
Supabase
ONNX Runtime
React 19

How Priority Scoring Works

Simple, explainable ML-powered formula

Priority Score =
(Bugs × 3 + Features × 2 + Competitors × 2.5 + Case Studies × 1.5)
× Urgency Multiplier × Churn Boost

📊Base Scores (Weighted)

  • 3.0xBugs: Severity × Frequency × Confidence
  • 2.0xFeatures: Priority × Competitor boost (1.5x if they have it)
  • 2.5xCompetitors: Threat level × Confidence
  • 1.5xCase Studies: Strength × Quote count

⚡Multipliers

Urgency Multiplier:
  • Low (1): 1.0x
  • Medium (2): 1.2x
  • High (3): 1.5x
  • Critical (4): 2.0x
Churn Boost:
  • At-risk customer (churn ≥ 2 & sentiment ≤ 4): +30%
  • Otherwise: No boost
🧠

ML Enhancement

The formula above is the fallback. In production, we use a trained GradientBoosting ML model (R² = 0.73) that learns from real call data to predict priority more accurately.

Method: ONNX Runtime
Accuracy: 73% variance explained
Features: 7 inputs
✅Transparent, explainable, and backed by ML

Fast Lane Slack Alerts

SA
Signal AI
APP · 2:34 PM
🚨 Fast Lane: High Priority Call Alert

Customer mentioned critical export bug affecting 3+ teams. Competitor comparison to Salesforce made.

Priority Score:42.5
Sentiment:4/10
🐛 Bug: Export feature not working (Severity: 4/4)
⚔️ Competitor: Salesforce (Threat: 3/3)

Instant Notifications

  • ✓High-priority calls trigger alerts
  • ✓Formatted with bugs, features, competitors
  • ✓Action items & key decisions included

Team Collaboration

  • ✓One-click "Send to Slack" from UI
  • ✓Automatic priority threshold routing
  • ✓Rich formatting with evidence

AI-Powered Customer Matching

Beyond transcription - true intelligence

1
Transcript
2
LLM Extract
3
Fuzzy Search
4
CRM Match
👤
Input
"Sarah from Acme Corp mentioned..."
→
Matched
Sarah Johnson, VP Sales @ Acme Corporation
Confidence: 94%
🏷️

No Manual Tagging

Auto-link calls to CRM records

🗺️

Journey Tracking

Follow customers across all touchpoints

⚠️

Risk Detection

Identify at-risk accounts early

Spot Trends Before
They Become Problems

Actionable intelligence, not just data

📈

Trend Analysis

  • ✓24h change detection
  • ✓Statistical significance testing
  • ✓New topics auto-identified
🎯

Call Clustering

  • ✓UMAP 2D visualization
  • ✓Group similar calls automatically
  • ✓Color-coded by signal type
⚔️

Competitor Intel

  • ✓Frequency tracking per competitor
  • ✓Threat level assessment
  • ✓Context extraction from calls
<10sDashboard Load Time

Why Signal AI Wins

FeatureManual ReviewGeneric AISignal AI
Speed✗ 15-20 min~ 2-3 min✓ <1 min ⚡
Accuracy✗ Variable~ ~60%✓ 90%+ ✓
Scalability✗ Limited~ Moderate✓ 360 req/min 🚀
Actionability✗ Low~ Medium✓ High 🎯
Cost✗ High (human)~ Medium✓ Free tier 💰
🏆Clear winner across all dimensions

Try Signal AI Now

Evidence + Volume = Action

🚀Live Demo📊Daily Signal Dashboard
Fast Lane
Try Now
Dashboard
View Insights
GitHub
View Code

VWO AI Hackathon 2025

Signal AI