ConceptNet

Voice intent classifier for enterprise AI agents

2 followers

Voice intent classifier for enterprise AI agents

2 followers

ConceptNet classifies enterprise voice commands into 4 intent layers — Basic, Context-Aware, Predictive, and Autonomous — producing structured JSON for agent execution. Token-free · 98.6% accuracy · 9 languages · No API cost · Open source Built for sales, legal, ops, and CS teams who want to automate repetitive workflows without paying $125K/year for GPT-4o. Try the live sandbox in 30 seconds — no login, no setup.

Free

Launch tags:Open Source•SaaS•Artificial Intelligence

Launch Team / Built With

Zendesk for StartupsUp to 2 years free for startups + AI Agents & Copilot

Promoted

Maker

📌

Hi Product Hunt 👋🏿 I'm Tony, founder of ConceptNet. I built this because enterprise AI agents have a classification problem. Every agent needs to know what kind of action to take — execute now, wait for a trigger, act proactively, or run autonomously forever. That's 4 fundamentally different execution patterns. No existing tool classifies them correctly. So I built one. Before raising a penny. What we have today: ✅ Fast-path classifier — 83% accuracy, <5ms ✅ Neural model — 98.6% accuracy, <100ms ✅ 730-example multilingual dataset ✅ 9 languages live ✅ Open source on GitHub ✅ Patents pending Looking for 20 enterprise teams to pilot free. Sales, legal, ops, CS if your team does repetitive voice workflows, this is for you. Try it in 30 seconds: conceptnet.co.uk/sandbox/ Happy to answer any questions below 👇🏿

Report

23d ago

Maker

Hi Product Hunt 👋🏿

I'm Tony — solo founder of ConceptNet.

I built this before raising a penny because I saw a gap nobody was talking about.

Every enterprise AI agent needs to classify what kind of action to take — not just what to do, but when.

Execute now? Wait for a trigger? Act before something happens? Run forever automatically?

Those are 4 completely different execution patterns. Nobody had defined them as a taxonomy. Nobody had built a classifier specifically for them.

So I did.

Something unexpected happened this week. An independent ML researcher from Hugging Face tried to break our model with adversarial holdout splits the kind of test designed to catch fake accuracy numbers.

He couldn't break it.

He confirmed 99.3% accuracy. He found 4 improvements. We fixed all 4 in 24 hours and retrained. Now hitting 100% on the standard test set.

What exists today — before raising any funding:
✅ Two-stage ML pipeline — 83% fast path, 100% neural
✅ 757-example multilingual dataset — 9 languages
✅ Live sandbox — try it right now, no login
✅ Open source GitHub
✅ Hugging Face model — 17 downloads already
✅ Patents pending

I'm looking for enterprise teams to pilot free — sales, legal, ops, government, BPO. If your team does repetitive voice workflows try it and tell me what you find.

Happy to answer any questions below 👇🏿

Hugging face
https://discuss.huggingface.co/t/conceptnet-4-layer-enterprise-voice-intent-classifier-98-6-accuracy-9-languages-token-free-open-source/179274

Sandbox
https://conceptnet.co.uk/sandbox/

Report

20d ago

Maker

Tested data sets 😉🙌🏿🙏🏿

Report

20d ago

Reviews

Hi Product Hunt 👋🏿

I'm Tony — solo founder of ConceptNet.

I built this before raising a penny because I saw a gap nobody was talking about.

Every enterprise AI agent needs to classify what kind of action to take — not just what to do, but when.

Execute now? Wait for a trigger? Act before something happens? Run forever automatically?

Those are 4 completely different execution patterns. Nobody had defined them as a taxonomy. Nobody had built a classifier specifically for them.

So I did.

He couldn't break it.

He confirmed 99.3% accuracy. He found 4 improvements. We fixed all 4 in 24 hours and retrained. Now hitting 100% on the standard test set.

I'm looking for enterprise teams to pilot free — sales, legal, ops, government, BPO. If your team does repetitive voice workflows try it and tell me what you find.