Skip to content

Applied AI – Tax Research

Why this role exists

Commenda helps companies operate across borders. We handle indirect tax, entity management, transfer pricing, and corporate tax and accounting for multinational businesses. We allow our customers to hire, sell, and operate in fifty jurisdictions without hiring fifty accountants.

The core challenge is that the law itself is fractally complex. To calculate a US sales tax rate, help a customer incorporate a Malaysian subsidiary, or determine whether they need to register for UAE corporate tax, we need structured, machine-readable models of all relevant laws. Our explicit goal is to schematize and index every law on earth that matters to our customers and to keep that index up-to-date.

This is much harder than it sounds. Tax law was not written to be schematized. There are 14,000 taxing jurisdictions in the US alone and each one can have tiered rates, partial exemptions that apply to only a slice of a transaction, and product categories that shift taxability based on price or use. Good schema design here is the difference between a system that handles the real world and one that quietly makes mistakes.

Customers use our data to move large amounts of money and make critical business decisions. It has to be correct and consistent with carefully-indexed sources.

It also has to stay correct. Laws change constantly, in thousands of jurisdictions, and no team of humans can track them all by hand. Our team’s AI-powered research systems monitor, research, and update our structured knowledge automatically, with humans and evals verifying that the machine got it right.

In addition to the law as written, the content team aims to model compliance outcomes as actually operationalized by government agencies. We process real registrations, filings, and notices at volume, which teaches us lessons that you can’t learn by reading a statute: how long an EIN application takes by fax this quarter, which state agencies are backlogged, or how a tax authority actually responds to a given notice type. Over time, this role will anonymize that internal data and index it into a structured knowledge base so that our ops teams, our product, and eventually our customers can benefit from what we've empirically learned about how the law is actually practiced.

The output of this work isn't just a database. It powers our tax engine, our filings, our dashboard and AI product, our marketing, and Commenda Media, our public research arm. Everything Commenda ships sits on top of what this team produces.

What you'll be accountable for

  • The schemas and their documentation. The design, correctness, and legibility of the content team’s core product. Every product at Commenda is built on top of this.
  • The automated systems that keep them current. AI research pipelines that detect legal changes, research them against primary sources, propose updates, and continually scan for mistakes. Freshness is an outcome you own. You will measure and optimize the time from when a law changes to when it’s reflected in our index.
  • Coverage, and the cost of expanding it. The team is constantly modeling new jurisdictions, new tax types, and new business domains. Your automation should relentlessly drive down the marginal cost of jurisdiction N+1: from weeks of human research toward hours of human review.
  • Correctness where it reaches the customer. No customer outcome should ever be wrong because the index was wrong, stale, or difficult to understand.

What you'll actually do

  • Schematize laws. Read statutes, regulations, and administrative guidance from primary sources and design schemas that capture what they actually say. Decide when a new jurisdiction's weirdness fits the existing model and when the model itself needs to change.
  • Automate the research. Build hosted, agentic research workflows: LLM pipelines that research a jurisdiction, populate a schema, and cross-check their own work. Think researcher/evaluator agent pairs, not one-shot prompts. Own the infrastructure these run on.
  • Evaluate and measure output relentlessly. Build eval suites, adversarial test cases, and fuzzing harnesses that measure whether the system is actually right. Design the ground-truth datasets. Know the difference between 99% accurate and correct.

What we're looking for

Required:

  • Extreme sense of ownership and urgency. Your work output is a schema and a pipeline, but your actual product is a customer’s compliance outcomes. When something goes wrong, you’re conscious of the lost money, time, and customer trust.
  • Software engineering background. You can build and operate real systems like data pipelines, APIs, and evaluation infrastructure.
  • Good at schema design. You have strong instincts for modeling messy real-world domains in structured form, and taste for when to generalize versus when to special-case. You can point to data models you've designed and defend the decisions.

Preferred, not required:

  • Experience using AI to automate knowledge work.
  • Background in tax, law, finance, or another detail-dense regulated domain

Where the role sits

You'll be part of a three-person team alongside the CTO and a research economist. It's a small team with an outsized blast radius: your output is consumed by every product team at Commenda.