What we're thinking about.
At Kepler we're building enterprise intelligence for high-stakes industries: AI that knows a firm's own data and proves every number. Code pulls, computes and cites each figure, and the model handles the question and the write-up. We started with investment firms, where a single wrong number can move millions.
Here we write about the ideas behind that work: how traceability changes what's possible with AI, why data provenance matters more than model size, and what we're learning as we build alongside the analysts and investors who use Kepler every day.
The Compiler That Never Says No
English is now a programming language whose compiler runs unclear instructions as readily as clear ones. Clear writing is the scarcest skill in technical work.
Read post
Your Mistakes Are a Dataset
When a system gets something right for the 10,000th time, you learn little. When it's wrong for the first time, you may have found what nobody knew to test for.
Kepler in Claude: A Working Guide for Analysts
Claude does the reasoning and Kepler does the retrieval and arithmetic, with every figure one click from the filing.
From Zero Context to Shipping
AI is collapsing the cost of understanding unfamiliar systems. The bottleneck moves to separating what you've observed from what you've inferred.
Two Forms of a Table
Three frontier models can't agree on the average of one column in a clean table. Why the form you read and the form you compute on differ, and why both matter.
The Real Painkiller for Financial Analysts: Kepler on the GameStop-eBay Deal
Using the GameStop-eBay bid, we run Kepler through four workflows that give analysts headaches, and show verifiable, accurate analysis in minutes, not days.
Elegant and Wrong
49.8% of an S-1's cells are empty. The clean data model you reach for first is a hypothesis the data can reject, and that shaped how we built citations.
How Quartr Helps Kepler Build Citation-Grade Earnings Research
How Kepler cites earnings answers across the SEC filing, the Quartr transcript, and the IR deck, with the language model never touching the numbers.
Why Data Pipelines Need Stronger Guardrails in the AI Era
AI can help you build faster than ever, but without the right guardrails it can silently erode trust and drive you off a cliff.
The Yardstick Problem
You can't unit-test a probability distribution. In our double-entry evals, code grades numbers, an LLM grades prose, and most value is where they disagree.
Eagerness solves most problems. Judgement solves the ones we work on.
Why building infrastructure for AI rewards pattern recognition over effort, and how that has reshaped how Kepler hires and operates.
FDEs in the Dark
The Forward Deployed Engineer was built to close the gap between software and reality. AI broke that model: an FDE without traceability can't find the answer.
How Parallel Helps Kepler Build AI That Finance Professionals Can Actually Trust
Kepler uses Parallel's Search API for discovery, pairing AI's grasp of open-ended questions with deterministic infrastructure for accuracy and auditability.
The Context Wall
Source-cited chat hits four walls: latency, cost, prompt-following degradation, and a context limit. Solving them meant rethinking what a conversation is.
Equity Research in the Age of AI: What Actually Needs to Change
Equity research is built on precision. AI doesn't need to replace judgment; it needs to remove operational drag and keep every number verifiably correct.
Context Is the Easy Part
Everyone's talking about context engineering. Context engineering isn't a context problem. It's an engineering problem.
Trust in the Age of AI
What does it mean to trust AI? Accuracy, security, and credulity are three related challenges, and forcing AI to show its work changes the output itself.
The AI Era of Finance: A Practitioner's View
LLMs excel at summarizing but struggle with the one thing finance can’t compromise on: numerical truth. Without traceability, AI risks accelerating errors.
Why I Joined Kepler
After Meta and four seed-stage startups, Susannah, Kepler’s Founding Engineer, shares why building fully traceable AI was a challenge she couldn’t pass up.
Introducing Kepler
We're building the infrastructure that gives AI a foundation of verified data, traceable to its source and precise enough to follow wherever it leads.