UdaraJay

UdaraJay / tiny-router

Public

Compact multi-head text classifier for short, domain-neutral routing decisions.

huggingface.cotgupjtiny-router intent-classification machine-learning nlp routing text-classification
13
2
100% credibility
Found Mar 24, 2026 at 13 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Tiny-router is a lightweight multi-task classifier that analyzes short texts with optional prior context to predict message relation, actionability, retention, and urgency for workflow automation.

How It Works

1
🔍 Find the message sorter

You discover a handy little tool that helps computers quickly decide what to do with incoming messages, like if they're urgent corrections or new ideas.

2
📚 Get ready examples

You use the included sample conversations, each marked with what kind of message it is, like a follow-up or something to act on right away.

3
🧠 Teach it patterns

You show the tool lots of these examples so it learns to spot message types on its own, getting smarter with each one.

4
📊 Check its guesses

You test it on held-back examples to see how well it understands relations, actions, importance, and timing.

5
💬 Sort a new message

You give it a fresh message with some chat history, and it instantly labels it with confidence scores for easy decisions.

6
📦 Pack it up lightly

You save the trained sorter in a tiny, fast format ready to drop into your own projects or apps.

🎉 Messages flow smartly

Now your system automatically prioritizes, routes, and remembers messages perfectly, making everything run smoothly without extra work.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 13 to 13 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is tiny-router?

tiny-router is a Python-based text classifier that routes short messages by predicting four signals: relation to previous context (new, correction, etc.), actionability (act, review, none), retention (remember, ephemeral), and urgency (high, low). It takes current text plus optional prior interaction details like recency and feeds them into small transformer encoders, spitting out calibrated JSON predictions for agent workflows or chat systems. Like a mini router tool for domain-neutral decisions, it handles inputs via CLI or snippets, with ONNX export for fast inference.

Why is it gaining traction?

It packs multi-head classification into a compact footprint using encoders like DeBERTa-v3-small or MiniLM, with full CLI support for training on JSONL data, eval metrics including F1 and calibration, and one-command ONNX quantization. Developers dig the temperature-scaled confidences for safe automation and ablations testing text-only vs full context modes. Stands out as a tiny router bit alternative to heavy LLMs for quick routing gates.

Who should use this?

Backend engineers building conversational agents need it for triaging user messages, like escalating urgent corrections or dropping ephemera. AI workflow devs prioritizing queues or managing memory in multi-turn chats will find the structured outputs plug right into automation pipelines. Ideal for prototyping support bots or mini router table setups where low latency matters.

Verdict

Grab it for lightweight routing experiments—solid CLI, docs, and ONNX make prototyping painless, but 13 stars and 1.0% credibility score signal early maturity without production tests or real datasets. Fine-tune on your data before deploying; it's a compact multi-head screwdriver worth sharpening.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.