Senior AI Engineer (Python,MSFT Foundry)

Acuity Analytics; West of England; On-site; GBP 120,000 - 160,000; Full time

Job summary

Acuity Analytics is seeking a senior AI engineer to take a Claude-based multi-agent prototype and elevate it to production-grade reliability for regulated environments.

You will define architecture, engineering standards, and mentor the team while delivering hands-on implementation of the orchestration, auditing, and data-layer integration across Postgres and Azure Data Lake Storage.

Qualifications

• Senior level (8+ years) with multiple systems shipped to production.
• Production Python (3.12): typed, packaged, tested code, clean structure.
• Claude-based agents: Claude Agent SDK or Claude Code, reliable long sessions.

Responsibilities

• Own the production architecture for the multi-agent system and defend choices with measurements.
• Harden the deterministic backbone around agents and implement failure handling, cost, and latency controls.
• Build evaluation harnesses and tests to score changes against reference results.
• Operate what you ship with monitoring, per-step cost and token accounting, incident handling and release notes.
• Move run state, metadata and outputs to Postgres and Azure Data Lake Storage with access control and audit trails.
• Establish standards: code review, CI/CD, branching, pre-commit, reproducible environments.

Skills

Senior level

Python

Claude

CI/CD

PostgreSQL

Azure Data Lake

Tools

Docker

GitHub Actions

Azure DevOps

Parquet

pyarrow

Job description

Ascent has recently been acquired by Acuity Analytics. This is both a significant milestone for us and a tremendous opportunity for you.

Acuity Analytics is a business with a strong global reputation, an impressive client base, and ambitious growth plans. We deliver deep insights and domain-led digital transformation to high-growth and heavily regulated organisations. To our customers, we bring a partnership that provides the talent, technology, and capability to enhance performance and operational efficiency.

About the role

Take a working Claude-based multi-agent prototype, which turns clinical study documents and raw data into validated analysis datasets and reports, and make it production grade: reliable, testable, auditable and ready for a regulated environment. You will make the architecture decisions, set the engineering standards and mentor the team while delivering hands-on.

Who we need

A senior AI engineer who has shipped many systems into production and owns the outcome: design, build, release, operate. We are not looking for a prototyper or a researcher. You have taken AI-based and data-heavy systems from first design to live use more than once, you know what breaks after launch, and you build so it doesn't.

Responsibilities

• Own the production architecture: decide how the multi-agent system is structured (orchestrator, parallel workers, independent checkers, repair loops) and defend those choices with measurements.
• Take systems from prototype to live: harden the deterministic backbone around the agents (step orchestration, status tracking, caching, validation gates, resumable runs), including failure handling, cost and latency control.
• Make quality provable: build evaluation harnesses and test layers so every change is scored against reference results before release.
• Operate what you ship: monitoring, per-step cost and token accounting, incident handling and release notes.
• Build the data layer: move run state, metadata and outputs to Postgres and Azure Data Lake Storage, with access control and audit trails.
• Set the standards: code review, CI/CD, branching, pre-commit and secret scanning, reproducible environments, validation documentation. Raise the bar for the engineers around you.
• Work with domain experts: turn the rules clinical programmers and statisticians write into precise, testable specifications.

Must have

• Senior level (typically 8+ years in software or ML engineering) with a track record of multiple systems shipped to production and run in live use. Be ready to walk us through them: what you built, what failed, what you changed.
• Production Python (3.12): typed, packaged, tested code (pytest, ruff, pre-commit), clean module structure, no notebook-only work.
• Agents built on Claude (Claude Agent SDK or Claude Code, tool use, prompt design, structured outputs), with experience keeping long agent sessions reliable and detecting when an agent stops early or invents output.
• Multi-agent systems in production: orchestration, parallelism, dependency ordering, independent verification, failure handling, cost control. Evidence from live systems, not only demos.
• Testing and evaluating LLM systems: fake or recorded backends, deterministic tests for the surrounding code, separate evals for model output.
• Git and CI/CD at team scale: pull-request workflow, GitHub Actions or Azure DevOps pipelines, Docker, versioned releases.
• Cloud and data: PostgreSQL (schema design, migrations, audit tables), Azure Data Lake Storage / Blob, Parquet, pandas/pyarrow.
• GxP delivery: you have built or supported software under validation: traceability from requirement to test, audit trails, change control, access control, documentation.

Nice to have

• CDISC SDTM/ADaM knowledge, or pharma/CRO analytics experience.
• Ability to read SAS or R clinical programs.
• Domino Data Lab or similar job-based compute platforms.
• MCP servers, long-document retrieval, prompt caching, agent observability and tracing.

Why Join Us

At Acuity Analytics, you'll have the opportunity to work across complex enterprise Azure environments while helping shape and improve our Managed Services capability.

You’ll solve challenging technical problems, influence standards and automation, mentor other engineers, and work closely with customers and technical teams.

People are at the Heart of our Business. By investing in people, we achieve exceptional results for our clients and create new opportunities for our teams to thrive.

Check out this page for more details.

For more details about Acuity Analytics please see here: Read here.