Founding AI Engineering in London

Founding AI Engineering

Concorde Resolution Ltd

London (hybrid) · Full-timeAbout Concorde

UK businesses write off billions every year in disputes they cannot afford to resolve. Between £1,000 and £250,000, a commercial dispute is too small to justify legal costs and far too large to absorb — so an unpaid invoice becomes a bad debt, and a workable relationship quietly ends.

Concorde resolves those disputes in days rather than months, for a fixed fee. Software runs the process end to end; a qualified mediator reviews and releases every outcome before it reaches either party.

Where we actually are

We are being deliberately straightforward about this.

Pre-revenue and pre-launch

The platform is in build now with WALT Labs, a Google Cloud Premier Partner. We are supported by Barclays Eagle Labs and LawtechUK, and we are raising a £500,000 pre-seed round.

Two people today

Founder and CTO. Dr Krzysztof Sielewicz, our CTO, holds a PhD in electronics, spent years at CERN building data-acquisition systems for the ALICE detector upgrade, and has fifteen years in ultra-low-latency and high-performance systems.

The start date depends on our round closing

We would rather say that now than waste your time. If you are in a stable role and need certainty next month, this is not yet the right conversation — but we would still like to talk.

The role

You take ownership of the AI platform at handover from our delivery partner and make it production-grade. You will be the first engineer, working directly with the CTO, with real architectural authority from day one.

The problems you would actually work on

They are the reason this job is interesting.

Cross-party fact reconciliation

Two parties submit evidence independently. The system extracts typed facts from each side and has to determine which refer to the same event — where "delivery completed 25 March" and "goods arrived 25/03 but incomplete" are the same event with different characterisations. Deterministic candidate generation, bounded model calls, four outcome states including we cannot tell.

Permission-filtered retrieval

Party A must never be able to reach Party B's private material. Filtering happens inside the store at query time, never on the results. A post-filter is a leak waiting for a bug, and this one is existential rather than merely important.

Five agents with hard limits

Intake, evidence, legal research, analysis, negotiation. An agent may propose; it may never publish. Everything lands behind a human release gate.

Citation integrity

Every legal proposition must resolve to a real retrievable source, verified automatically before it reaches the review queue. Fabricated authority is the known failure mode in this domain and we treat an unresolvable citation as a blocking defect.

Genuine multi-model architecture

Vertex AI Model Garden, at least two model families in production, switching a task is a config change plus an evaluation run — never a code change.

What we are looking for

• Strong Python, and production experience with LLM systems — not demos
• Multi-agent orchestration: retries, idempotency, resumability, budget ceilings
• RAG done properly, with a real view on permissions and retrieval quality
• Google Cloud, or the ability to get there quickly
• Evaluation as a discipline — you should find "we tested it manually" unsatisfying
• Comfort with ambiguity, and the judgement to know when a problem needs a model and when it needs code

Not required

legal background. You will learn the domain.

Compensation

TBD plus meaningful equity .

We will be direct: that salary is below the London market for this role. We know it. The equity is real, the technical ownership is real, and you would be the first engineer on a platform being buil. If alot of cash is your priority right now, we understand — and this is not the role.

Concorde Resolution Ltd, No. , England & Wales. Concorde is not a law firm and does not provide legal advice.

#J-18808-Ljbffr