Senior Software Engineer, AI Agents
Confirmed on the employer's own hiring board on Nov 10, 2026. First seen by Alion on Nov 10, 2026.
We’re building agents that take on drafting work, starting with the routine parts, and we’re hiring a senior engineer to help build them.
About DocuSketch
DocuSketch helps restoration professionals document damaged properties and turn what they capture into floor plans and repair estimates. The platform supports tens of thousands of claims a month, and behind each floor plan is a drafter working from 360° imagery of the property.
The role
Our agent work spans floor-plan drafting, review, resolving differences between preliminary and detailed plans, and triaging incoming orders, using 360° imagery and geometry where relevant. Drafting is the company’s core service, so the value of each agent is direct.
You’ll join the agentic team, report to the Agentic Team Lead, and own agent capabilities from design through evaluation and rollout, working with the AI team that trains our vision models and with experienced drafters. This is a hands-on engineering role.
Where you start
Our first agents are moving into production. Much of the next work is engineering: running agents reliably at scale and bringing their output into drafters’ daily work, while extending what they can do.
For each new capability, you’ll answer the same questions, with evidence: What should the agent see, and which tools and data does it get? Where is a model the right answer, and where is ordinary code? How will you know it works before a drafter relies on it?
What you’ll do
- Choose what to automate. Work with experienced drafters to find the steps and mistakes that cost the most, define the evidence an agent needs for each, and present its output so a drafter can check and correct it.
- Design each agent’s tools and strategy. Choose models, decide what to expose from the structured floor-plan data and which imagery to supply, and work out where model reasoning, geometry or ordinary code gives the best answer.
- Own the evaluations. Build them from real properties and expert drafting and review. Measure the errors that matter for each task, such as missed defects and false alarms for a check, compare approaches, and say plainly what a result does and does not show. These results decide when a capability is ready for wider use.
- Put agents into the drafting workflow. Write the Python services and integrations that deliver their output, debug across service boundaries, and make practical trade-offs in cost, latency and reliability.
- Turn failures into fixes. When an agent gets a plan wrong, misses a defect or raises a false alarm, read the trace, find the cause, add the case to the evaluation set and change what needs changing.
What we look for
- 4+ years of hands-on technical work, with Python as a working language; relevant years in military or national-service technology units count. We particularly value machine learning, data science and deep learning backgrounds, and welcome backend engineers who build agents; much of the near-term work is production services and infrastructure.
- You build agents, at work or on your own time. You can explain the choices you made about their tools, context and execution, and how you established that they worked. The field is young, so we look for depth and curiosity rather than years.
- Evaluation judgment. You can turn an observed failure into a useful test case, tell a meaningful measure from a flattering one, and work with domain experts when correctness takes judgment.
- Initiative. You find the important problem, cut it down to something you can ship, and follow it into real use. You’d enjoy working like an early engineer in a small team inside an established company.
- Fluency with coding agents. You use them to build, test and debug software, and you can inspect, explain and take responsibility for the code you ship.
Experience with visual reasoning, multimodal models or geometry is useful. You can learn the restoration domain here, and we don’t expect experience with any particular agent framework.
Why join us
- You work with experienced drafters who check agent output and help trace errors.
- A profitable company.
- Fully remote, with relocation to Germany possible.
- AI-forward, with generous token budgets.
- Flexible employment arrangements.
- Working language: English.
How we hire
- Interview with HR
- Short interview with the hiring manager
- Take-home assignment
- Deep, hands-on technical interview: we build agents
- Interview with an executive
How to apply
Apply through this page and tell us about an agent you built: what it had to do, an engineering decision that changed what it could do, and how you knew the change worked. A clear account of your work is enough. You can describe proprietary work without sharing anything confidential, and a public repository is optional.
We welcome applicants of all genders and backgrounds.