When a soldier is wounded, the first record of their care is often a note on a card, a radio call or a medic's memory. Much of it is lost at the first handover. Caslog replaces that with a small recorder that clips onto the casualty, listens to the care being given and turns it into a clinical record that follows the patient from point of injury to hospital. We build it together with combat medics in Ukraine.
We are two founders, Swiss and Ukrainian, both serving military officers, and we have closed our pre-seed round. We're looking for an all-round engineer who builds with agents better than we do, and who makes the whole team faster at it.
WHAT YOU'LL WORK ON
Whatever moves the product forward most, across the whole stack. At the start that is mostly the AI core:
The speech and extraction pipeline: noisy field audio in, a structured clinical record out, which a human reviews and signs off.
The improvement loop: every change measured against the last one before it ships, on what medics care about (missed events, wrong details, invented content).
Backend, security and infrastructure: encryption, access control, a full audit trail, processing that survives crashes, deployments customers can run themselves.
How we build with agents: the harness, context, review and tests that let a very small team ship what used to take a department.
Your first goal: get the pipeline ready for the device's field tests with Ukrainian medics, with an evaluation set agreed with medics and results that measurably improve against it. You work closely with our technical founder and shape the backlog together. Good arguments change the plan.
OUR STACK
Python, PostgreSQL, Docker and GitLab CI on Google Cloud. Model work in PyTorch, Hugging Face Transformers and NVIDIA NeMo where the product needs it. If you'd pick something better for a job, say so.
WHO YOU ARE
You are a top agentic development talent. You use Claude Code, Codex, OpenCode, Hermes or a similar harness every day, and you know it better than most people who build with it. You can show, for example, that you:
run several agents in parallel on isolated branches or worktrees and merge their work without chaos;
write the specs, tests and checks that let an agent verify its own work, and catch it when it says "done" without having run them;
keep context lean and durable: project instruction files, skills, subagents, hooks, MCP tools and notes that survive between sessions;
put limits around agents: sandboxes, least-privilege permissions, review gates, token and spend budgets;
evaluate the agents themselves: compare models and prompts on your own tasks, find where they fail and fix that in the process, not only in the prompt;
use agents beyond code: labelling data, generating synthetic data, triaging logs and incidents;
leverage local models and keep up with recent developments such as Decisions APIs.
You have 5+ years of building and running production software, so you know what good code looks like and notice when an agent gets it wrong. Underneath, you're a real engineer:
You have depth in backends, and you've fixed them when they broke in production.
You treat data as carefully as code and think about security by default: clean test splits, authentication, secrets, least privilege.
You are pedantic about cleaning up repos and keeping them tidy, simple and secure.
You can design a system end to end and explain your choices in plain words.
Bonus: fine-tuning and serving models, audio and speech work, speaking Ukrainian, a background in medicine or the military.
FORMAT AND WHAT WE OFFER
Full-time, part-time or contractor: we're open to the format that fits you. Pay and share options depend on the format and the time you commit, and we agree them.
Work on the core of a product with an unusually clear purpose: wounded people getting the right care.
We pay for your coding-agent subscriptions. Fully remote, with occasional meetups in Kyiv.
This is defense
