/research

Insights and blogs from the AI Group building Fin at Intercom

What Does It Mean for a Model to ‘Think’? Reasoning, Recursion, and the Operator Design Space

Looped Transformers, thoughts on how LLMs Reason, and an overview of a recent paper we wrote.

Read

Articles

23

Structured, Agentic RAG for Ecommerce

2026.05.07

Low-Rank Key Value Attention: Reducing KV Cache Memory and Maintaining Head Diversity

2026.04.09

Unsupervised Learning Meets Generative AI: Topic Modelling for Real-World Dialogue

2026.03.30

Podcast EP2: Shipping reliable AI actions

2025.09.19

Podcast EP1: Closing the loop

2025.09.18

How We Built a World-Class Reranker for Fin

2025.09.11

Using LLMs as a Reranker for RAG: A Practical Guide

,

2025.09.11

Finetuning Retrieval for Fin

2025.09.11

David vs Goliath: are small LLMs any good?

,

2025.09.11

Building out Intercom’s AI infra

2025.09.11

“Was that helpful?” Understanding User Feedback in Customer Support AI Agents

2025.09.11

To escalate, or not to escalate, that is the question

2025.09.11

Building a Better Language Detection Model for Fin

2025.09.11

Cost of Serving LLMs

2025.09.10

We Don’t Need Higher Peak Intelligence, Only More Intelligence Density

2025.08.20

Fin: Running a Reliable Service over Unreliable Parts

2025.08.08

Think Fast: Reasoning at 3ms a Token

2025.07.18

A Causal Inference Approach to Measuring the Impact of Improved RAG Content

2025.07.07

Generating Knowledge Center Content from Customer Service Conversations

2025.07.07

Do you really need a Vector Search Database?

2025.04.29

The Agency, Control, Reliability (ACR) Tradeoff for Agents

2025.04.11

An Actor-Critic Approach to Reduce Hallucinations

2025.04.10

Slower Feels Smarter? Experimenting with AI Agent Latency

2025.04.10