"Gauntlet team on AI Infra Summit Hackathon Hackathon"

Team Idea

Gauntlet a load, chaos and cost benchmark rig for real-time voice agents. Voice agents ship on vibes: teams demo one clean call and have no idea what happens at 50 concurrent callers, on a lossy network, or when the user interrupts mid-sentence. Gauntlet is the missing test harness. It spawns synthetic adversarial callers over WebRTC (accents, cross-talk, packet loss, barge-in), measures true acoustic round-trip latency and interruption handling, scores transcript accuracy against an independent referee STT, sweeps model/STT/TTS configs to produce a cost-vs-latency-vs-quality Pareto frontier, and exposes a CI gate that fails the build when a regression slips in. Building solo not looking for teammates, but happy to trade notes with other voice and infra teams.

Team Members

Hamza Ahmad

hamzaahmad3006

AI Engineer

Submission

GAUNTLET: Crash-Test Rig for Real-Time Voice AI

GAUNTLET dials voice AI agents with synthetic callers, measures latency and turn-taking from the caller's ear under seeded network chaos, and returns a deterministic readiness score that can fail a pull request.