AI Engineer

Filed is building the first AI tax preparer trusted by America’s accounting firms.

This industry might seem boring—but it’s a $70B backbone of the economy that’s breaking under a talent shortage no one has solved. We’re fixing it—fast.

In just 9 months, we’ve hit strong product-market fit, are generating insanely fast-growing, ridiculously sticky revenue, and are backed by top-tier investors. We’re not a tool—we’re the missing workforce firms have been sounding the alarm for. We’re building the future by expanding our already 30-person team with those who want to say “I was there when it all started.”

Read more TechCrunch

⸻

What we should tell you about the role:

We're looking for an AI Engineer who moves fast and thrives in high-velocity environments. You're the kind of person who sees a model failure mode, understands the root cause quickly, and ships a new eval, prompt, or pipeline in hours, not weeks. You love turning messy, unstructured tax data into working systems, and you don't wait around for perfect inputs to get started.

You're a builder who's just as comfortable designing agentic workflows and retrieval pipelines as you are wiring up evals, tracing failures, or debugging a flaky tool call. You sweat the details, but never let them block momentum. You think in terms of accuracy, latency, and iteration, always asking: how can we get this in front of users faster, and how do we know it actually worked?

As one of our core AI hires, you'll help shape how Filed reasons about tax returns end to end. You'll work closely with product, tax engineering, and most importantly, our users, to rethink how AI can do real, accountable work in a space most people thought couldn't be automated.

⸻

Here’s why this is a terrible job:

• This is not a job for someone who likes to move slow.
• We operate week to week with ruthless speed. Sometimes we're going in the wrong direction, but more times than not we're going in the right direction.
• If you need a few extra days to perfect a pipeline, chances are there's already a handful of team members blocked behind you.
• We use tools like Claude, OpenAI, Pydantic AI, and our own eval infra to move fast, test ideas in days, and ship to production quickly.
• You'll be operating at the intersection of LLMs, tax domain logic, and high-stakes accuracy requirements.
• There are no clear answers. Just judgment calls, ambiguity, and constant input from every direction.
• You'll ask yourself, is this really the work? It is. And it's not for everyone.

⸻

Here’s why it’s the best career choice you’ll make:

• We're building what tax pros call impossible. A product that behaves like a teammate, not just another tool.
• We run reviews every 4 months. We operate with total transparency around pay, performance, and progress.
• You'll be paid above market. You'll earn equity that grows fast. And you'll have full autonomy to make high-impact decisions.
• You'll ship models and agents because you believe in them, not because they got signed off.
• You'll talk to customers weekly. You'll drive outcomes end to end.
• We'll tell you what needs to be achieved. How you get there is on you.
• We expect you to build a tax Oracle. An internal evaluation system so dialed in that you can predict exactly where the model will fail before a CPA ever sees it.
• That comes from immersion, curiosity, and care.

⸻

The teammate we’re looking for:

• Thrives at solving operational inefficiencies and systems challenges
• Moves seamlessly between identifying problems, implementing solutions, and iterating quickly
• Excited by the challenge of making internal operations invisible (because they just work)
• No lone wolves — we need a collaborative operator
• Big bonus if you have experience implementing automation or AI agents in a work environment

⸻

What you’ll be responsible for:

• Building and maintaining best-in-class LLM pipelines and agentic workflows in production
• Designing evals that catch regressions before users do, and grading systems that measure what actually matters
• Owning the retrieval, context, and tool-use infrastructure that powers Filed's AI tax preparer
• Driving accuracy, latency, and cost improvements through experiments, fine-tuning, and prompt engineering
• Enabling the broader engineering and tax teams with tooling, primitives, and shared infrastructure

⸻

Your hard skillset:

• Comfortable designing and analyzing evals, traces, and failure modes at scale
• Confident working across the stack (Python, TypeScript, vector stores, orchestration frameworks, observability)
• Bonus if you've worked in fast-paced startups or regulated industries
• Even bigger bonus if you've taken an AI product from 0 1 and watched real users depend on it

⸻

Your soft skillset:

• Comfortable being uncomfortable and making calls with incomplete data
• Empathetic and curious. Building for the user, not just for benchmark scores
• Highly collaborative across product, engineering, and tax
• Detail-oriented but pragmatic. You know when accuracy matters and when to ship
• You own your work and welcome the accountability that comes with it

⸻

Growth and Levels:

• We operate on six levels, from Level 1 to Level 6. Each reflects increasing ownership, impact, and scope with meaningful jumps in salary and equity.
• We review levels every 4 months. Promotions are based on output and ownership, not tenure.
• This isn't a comfy AI research role. It's fast, focused, and customer-obsessed.
• You will launch fast. You will learn fast. You will own what you ship.

⸻

The process:

• Quick intro call (30 minutes)
• Technical walkthrough and async task (60 minutes)
• Final round with our team (120 minutes)

In just 3.5 hours, you’ll be wanting early access - just like the rest of our team - to get into the stack and start building a generational company, even before your first day. Because you’ll see the opportunity to change your career trajectory - fast.