AI Application Engineer

𝐀𝐛𝐨𝐮𝐭 𝐕𝐞𝐫𝐢𝐝𝐨𝐱 Veridox works with some of the largest insurers, airlines, law firms, and financial institutions in the world to detect fraud in documents and images. When a customer submits a receipt, ID document, or damage photograph, our platform determines whether it is genuine, tampered with, or AI-generated and, critically, explains why with clear, verifiable evidence. We are building the production system behind that: it takes unstructured documents from enterprise clients and returns structured, human-actionable output. We're scaling rapidly, and we hold ourselves to a high standard because our customers act on what we tell them. 𝐖𝐡𝐞𝐫𝐞 𝐲𝐨𝐮'𝐥𝐥 𝐟𝐢𝐭 𝐢𝐧  We're looking for an experienced engineer to work across the whole of that system and make it better. You'll have real ownership of it, and a team alongside you who care about it as much as you do. 𝐖𝐡𝐚𝐭 𝐲𝐨𝐮'𝐥𝐥 𝐝𝐨 - Work across every stage of the pipeline: ingestion and parsing, classification and routing, the LLM analysis steps, third-party data integrations, and the deterministic logic that assembles the final output. - Take work end to end. You'll design it, build it, ship it, and then watch how it behaves in production. - Improve accuracy wherever it's weakest, which means being equally willing to fix an OCR problem, restructure a prompt, add a data source, or replace a model call with plain code. - Integrate external data providers and APIs, and handle their failure modes honestly, so that a missing answer never reads as a confident one. - Extend the evaluation and testing around your own work, so changes can be proven rather than asserted. - Support per-client configuration and onboarding without it turning into bespoke forks. - Work directly with the people who consume the output. Our results are read and acted on by humans, and we think engineers should hear from them. 𝐖𝐡𝐚𝐭 𝐰𝐞'𝐫𝐞 𝐥𝐨𝐨𝐤𝐢𝐧𝐠 𝐟𝐨𝐫 - 5+ years of professional engineering experience, with recent years spent on LLM systems in production. - Genuine breadth. You're comfortable moving between data pipelines, prompt and model work, API integrations, and backend service code in the same week. - Strong production Python. - Sound judgement on what belongs in a language model and what belongs in deterministic code. You'll have a clear view on where that line sits, and you'll make the case for it. - Evaluation experience. You've built labelled test sets, written the guidelines, and had a metric block a release, so you know what it takes to prove a change is an improvement. - Intellectual honesty about your own results. Our product exists because being confidently wrong is expensive, and we hold our own work to the same standard. We want someone who tries to break their own numbers before anyone else does. - The instinct to debug upstream-first. You look at what the model was actually given before you touch the prompt. - Initiative. You spot what needs doing and bring new ideas to the table without waiting for a backlog. Where the definition of "good enough" doesn't exist yet, you'll propose one and help us agree on it. - Strong written and verbal communication. - Rapid development 𝐍𝐢𝐜𝐞 𝐭𝐨 𝐡𝐚𝐯𝐞 - Document AI: OCR, VLMs, layout, classification and extraction at production quality. - Experience with systems where inputs can't be trusted, or anywhere being confidently wrong is expensive. - Experience working with noisy or imperfect ground truth. You've had labels that turned out to be wrong, and you know how to tell a model error from a bad label. Time spent in annotation and eval tooling of any kind (Label Studio, W\&B/Weave, Opik or similar) is useful background here, though none of it is our stack. - Experience where you personally held the quality bar for a production ML system. 𝐇𝐨𝐰 𝐰𝐞 𝐰𝐨𝐫𝐤 We are, first and foremost, builders. The team is small, mature, and fully remote, and we place real weight on humility, collaboration, and accountability. We solve the hard problems together and share the credit for them. - We move fast. Speed matters here. We favour shipping something good, learning from the real world, and improving it quickly over pursuing perfection before anything sees production. We expect people to make good decisions, deliver quickly, and keep iterating. Momentum matters. - Autonomy with support. We hire experts and trust them. You'll own work from concept through to production, backed by a team that has your back when it gets difficult. - Work that's read by people. Our output isn't a dashboard nobody opens. It's read and acted on by analysts at some of the largest enterprises in the world. - No ceremony. We keep process to what's genuinely useful so you can spend your time on the system rather than around it. - Monthly hackathons in Manchester/London locations