IntentiAI

IntentiAI is an intent classification engine for AI agents and chatbots. It doesn't just route to cheap vs expensive models. It understands what the user wants, then selects both the optimal model AND a tailored prompt for that specific intent.

How it works:

1. User query hits IntentiAI (<20ms classification on CPU)
2. Intent is identified (e.g. "billing inquiry" vs "technical troubleshooting")
3. Each intent maps to: the right model + an intent-optimized prompt

• "Check my balance" ? fast model + short factual prompt ? instant, accurate
• "Help me plan my retirement" ? powerful model + deep reasoning prompt ? thorough, precise
• Off-topic spam ? blocked entirely (zero cost, zero risk)

4. The AI gives better answers because it's specialized per intent, not generic

Why this is better than one-model-fits-all:

• Better accuracy: each intent gets a prompt engineered for that specific task
• Lower cost: simple intents don't need expensive models
• Faster responses: lightweight intents get lightweight models
• Built-in guardrails: off-topic queries never reach any model

Results:

• 91% F1 accuracy on BANKING77 benchmark (77 intent classes)
• 94% out-of-scope detection recall
• 70% average cost reduction
• No GPU required. It runs on CPU

Backed by published research. Free tier available with 2,000 queries/month.