TachyRoute
The Explainable, Multimodal, Early-Exit Decision Engine
Start new thread

TachyRoute - The Explainable, Multimodal, Early-Exit Decision Engine

by•
An open-source Non-Autoregressive Decision Engine. It evaluates typed decisions (choice, score, boolean) in <15ms using adaptive early-exit compute and attention rollout evidence extraction: without generating text, hallucinating or parsing fragile JSON.

Add a comment

Replies

Best
Maker
📌
Hey Product Hunt! 👋 I'm Dhanush, the creator of TachyRoute. We rely on Massive LLMs for everything, but why run a 24-layer transformer for a simple routing question? It's slow, expensive, and opaque. I built TachyRoute to fix this. It is a Non-Autoregressive Decision Engine that introduces the "TachyRoute Leap": ⚡ Adaptive Early-Exit Computing: If the model is confident at layer 6, it exits. P99 latency drops to ~15ms while retaining 100% accuracy. 🔍 Explainable Evidence: It uses Attention Rollout to return the exact text span that caused the decision, adding 0ms to inference time. 🔀 Dynamic Complexity Routing: Seamlessly redirects workloads between lightweight models and massive multi-lingual experts based purely on necessity. You can evaluate typed decisions over any state in milliseconds: without hallucinating or parsing fragile JSON. TachyRoute is fully open-source. Check out the GitHub repo, run the quickstart and let me know your thoughts or questions below!