Applied AI Engineer Remote (US East)

Weve Security Limited; United States; Remote; USD 120,000 - 180,000; Full time

Job summary

Weve Security Limited is building workplace security that connects activity, identity and business context across multiple systems. You’ll join an applied engineering team that brings research into production, balancing model quality with latency and cost.

You’ll work with security researchers and back-end engineers to implement robust investigations, with a focus on traceability, evidence gathering, and responsible AI use. This role is remote in the United States.

Qualifications

• Experience shipping AI/ML-backed features and assessing their behaviour with real users or production data.
• Strong software engineering skills and ability to build reliable data, model and tool integrations.
• Hands-on experience with LLM tool use, retrieval or multi-step reasoning, including failure modes.

Responsibilities

• Build investigation agents that gather evidence across connected systems and retain traceable support for conclusions.
• Design retrieval and context assembly around identities, activity and resources, including conflicting records and incomplete information.
• Evaluate lightweight models, custom ML and frontier models for different tasks, considering quality, latency and cost.
• Create repeatable evaluations for factual support, false escalation and robustness to noisy inputs.
• Implement safeguards around tool permissions, sensitive data and untrusted content encountered during investigations.
• Monitor production behaviour and drive improvements from failure analyses.

Skills

AI/ML features

Software engineering

LLM tooling & retrieval

Experimentation

Trade-offs (accuracy/latency/cost)

Tools

LLM tooling

Job description

Weve is building workplace security for a world where people and AI agents work across dozens of systems. We connect activity, identity and business context so security teams can investigate risks that no single tool can explain on its own.

An investigation can sound persuasive and still be wrong. It may join the wrong identities, overlook a benign explanation or cite a record that does not support its conclusion. You’ll build the retrieval, reasoning and evaluation machinery that helps us catch those failures.

This is applied engineering with a research component. You’ll take approaches from experiment to production, working with security researchers and backend engineers. Model quality, evidence quality, latency and cost all have to hold up outside a curated demo.

What you’ll own

• Build investigation agents that gather evidence across connected systems, choose appropriate tools and retain traceable support for their conclusions.
• Design retrieval and context assembly around identities, activity and resources, including conflicting records and incomplete information.
• Evaluate lightweight models, custom machine learning and frontier models for different tasks. Choose based on measured quality, latency and cost.
• Create repeatable evaluations for factual support, false escalation, missed evidence and robustness to noisy or adversarial inputs.
• Implement safeguards around tool permissions, sensitive data and untrusted content encountered during an investigation.
• Monitor production behaviour and work through failures with researchers and field engineers. Turn failure analysis into better datasets, tools and system design.

What you’ll bring

• Experience shipping AI or ML-backed features and assessing their behaviour with real users or production data.
• Strong software engineering skills and the ability to build reliable data, model and tool integrations.
• Hands-on experience with LLM tool use, retrieval or multi-step reasoning, including the ways these systems fail.
• An experimental approach: establish a baseline, isolate a change, evaluate representative cases and inspect regressions.
• Comfort making trade-offs between accuracy, latency, compute cost and operational complexity.

Useful, but not required

Security investigations, behavioural modelling, prompt-injection defence, smaller-model optimisation or experience with durable workflow systems.

What good looks like

• Changes improve a defined investigation task against a baseline, with regressions visible before rollout.
• A reviewer can trace important conclusions to evidence and recognise when the system does not have enough information.
• Useful investigation depth remains practical at customer workloads, with understood cost and latency.

Working here

You’ll work directly with the people building and selling Weve. We expect you to question the brief, make decisions within your area and follow through after something ships. Customer evidence matters more than how convincing an idea sounds in a meeting.

We work remotely and write down our decisions so people can make progress without waiting for everyone to be online. Read more about the team and why we’re building Weve .