Software Engineer Intern

Invariant AI
San Francisco, California, United StatesFull-timePosted Sep 18, 2026

About the role

Software Engineer, Intern

Location: San Francisco, CA, on-site Type: Internship, six months, full time Start: September 2026

About Invariant

Invariant AI builds autonomous agents that accelerate regulatory compliance for mission-critical industry. Our engineers deploy into the customer’s team and own the regulatory process end to end, from the first determination through to filing, with the agents doing the work underneath.

The problem is specific. A satellite operator gets the hardware built and then waits on paper. An FCC Part 25 grant for spectrum. A Part 5 experimental licence for the test campaign before that. An FAA Part 450 licence for the launch. A NOAA licence if a camera points down. An ITAR or EAR determination before a single component crosses a border. Each one runs to thousands of pages of narrative, orbital debris analysis, link budgets and citations into rules drafted before anyone had heard of a 100-satellite constellation. A lead systems engineer loses something like 18 hours a week to it, and outside counsel still bills six figures per spacecraft. Our customers get that down to three or four hours a week and reach the pad sooner.

About the role

You’ll join the engineering team as a software engineer intern and work on the agents that draft, check and file regulatory submissions. The work is production engineering, not research: the code you write ships to customers, and it ships on a measured improvement against an eval, not on plausibility.

You’ll work directly with the CTO, take part in customer working sessions, and own features end to end from the first conversation to the deployed change.

Responsibilities

Build and maintain agent workflows that read a regulatory corpus and draft filing sections with citations that resolve back to the rule text

Write behavioural evals before the feature, and use them to decide whether a change ships

Improve retrieval over large rule corpora: chunking, embeddings in pgvector, reranking

Run the agents on real customer data, review the output, and turn what you find into concrete fixes: prompts, eval cases, workflow steps

Turn customer and reviewer feedback into shipped changes between one working session and the next

Build production Python services: FastAPI, background workers, streaming progress into a Next.js interface

Handle failure modes and refusal behaviour, and decide where the agent should stop and hand back to a human

Requirements

Studying, or recently finished, computer science or a close neighbour. Or you skipped that route and have the work to show instead, which counts equally

You’ve built something with LLMs that survived contact with a real user. A notebook doesn’t count

Strong Python. Comfortable with async, type hints, and a codebase where mypy --strict is a merge gate

Enough statistics to tell a real improvement from noise on a small eval set

Careful with sensitive material

Six months, full time, in person in San Francisco. Not alongside a course load

Nice to have

A cubesat team, rocketry club, satellite ground station, or a stint at a space agency or operator

Coursework in RF, orbital mechanics, or guidance and control

Retrieval work deep enough that you know when reranking helps and when it does nothing

Open source contributions we can read

Compensation and what you get out of it

Competitive Pay

A full-time offer at the end for interns who earn one

Direct work with both founders. There’s no layer between you and the decisions

Your name on work that ships to companies putting hardware in orbit

Equal opportunity

Invariant AI is an equal opportunity employer. We do not discriminate on the basis of race, colour, religion, sex, sexual orientation, gender identity, gender expression, national origin, ancestry, citizenship or immigration status, age, disability, medical condition, marital status, veteran status, or any other characteristic protected by federal law, by California’s Fair Employment and Housing Act, or by San Francisco ordinance.

How to apply

Direct apply or email jobs@invariant-ai.com with the subject line “Software Engineer Intern”. Skip the cover letter. Send us one thing you’ve built with an agent loop, a repository or a link, plus one paragraph on what it got wrong and what you did about it.

Responsibilities

  • Build and maintain agent workflows that read a regulatory corpus and draft filing sections with citations
  • Write behavioural evals before the feature and use them to decide whether a change ships
  • Improve retrieval over large rule corpora
  • Run the agents on real customer data, review the output, and turn findings into concrete fixes
  • Turn customer and reviewer feedback into shipped changes
  • Build production Python services

Qualifications

  • Studying or recently finished computer science or related field
  • Experience building something with LLMs
  • Strong Python skills, comfortable with async and type hints
  • Understanding of statistics for evaluating improvements

Benefits

  • Competitive Pay
  • Full-time offer at the end for successful interns
  • Direct work with founders

Skills mentioned

PythonFastAPINext.jsPostgreSQLGenerative AILarge Language ModelsRetrieval-Augmented GenerationAI AgentsEmbeddingsModel Evaluation

About Invariant AI

At Invariant AI, we want to enable the most ambitious deep tech founders to completely focus on engineering and leave the cumbersome paperwork, compliance, insurance, licensing and all things back office to us. We provide engineers and consultants who are equipped with our AI platform and take care of all your FCC, ITU, NRC, WPC filings, ground station setups, ERP creation and all things back office.

Technology2-10 employeesSan Francisco, California