Intercom AI

40+ machine learning scientists, engineers, and designers building Fin — the #1-rated AI agent for customer service.

Fin AI Engine

Fin is powered by a proprietary, patent-pending AI Engine with six optimization layers that work together to resolve customer support conversations across chat, email, voice, and social channels:

1. Query Refinement — reformulates and contextualizes customer questions
2. Content Retrieval — surfaces relevant knowledge using our fin-cx-retrieval model
3. Reranking — precision-ranks candidates with our fin-cx-reranker model
4. Response Generation — produces answers via Fin Apex 1.0, our core generative model
5. Accuracy Validation — checks responses against safety and correctness standards
6. Engine Optimization — continuously calibrates the full pipeline

Research Areas

Our team works at the intersection of applied ML and production-scale customer service:

• Retrieval & Reranking — domain-specific embedding and cross-encoder models for enterprise knowledge bases
• Agentic Tool Use — procedure execution, API orchestration, and multi-step reasoning
• Fine-Tuning & Alignment — GRPO training and supervised fine-tuning for customer support domains
• Evaluation & Safety — hallucination detection, answer quality scoring, and automated red-teaming
• Dialogue Understanding — topic modelling combining unsupervised learning with generative AI for conversation analysis and continuous improvement

📄 Read more on our AI Research Blog

Links

• 🌐 fin.ai
• 🔬 Research
• 🐙 GitHub
• 🐦 Twitter
• 💼 LinkedIn