ENNPRO Get in touch
← All engineering roles

Principal AI Solutions Architect

Design the handoff between a client’s AI service and the operators responsible for it.

Apply for this role

About the role

You’ll work with a client’s service owners to map where an AI component belongs in an existing workflow. Your design has to account for the request that times out, the record that changes, and the decision that still needs a person.

You’ll work across the client’s application, data and operations teams. A design review should settle who owns each dependency and what a caller can expect when that dependency is slow, unavailable or working from an old record. You’ll prototype the uncertain parts yourself.

Working at ENNPRO

ENNPRO is an applied machine learning consultancy. We work alongside client engineers and the people who use their systems. Depending on the engagement, a client might need a service they can operate, a benchmark they can reproduce or enough evidence to decide whether to release a model. Their team should be able to use the result after handover.

You’ll be the person who brings the client’s application owner, platform engineers, data owners and operations lead to the same design decision. Each has a different view of the service. A product owner sees the task a user is trying to complete. A platform engineer sees an extra dependency in a production request. A reviewer sees the record that needs an explanation six months later. You’ll represent all three views in a design the teams can implement.

This is a principal individual-contributor role with substantial hands-on work. You may write a small adapter yourself, pair with a client engineer on a release rehearsal or prepare a decision for a sponsor who needs to understand the cost of changing direction. You’ll give senior engineers room to own their components and resolve the issues that cut across them.

Responsibilities

  • Follow a representative request through the existing workflow. Identify the system of record, access decisions, latency budget and point where a person commits an action.
  • Specify service and event contracts. Work through duplicate requests, retries and changes to source records with the engineers who own the surrounding systems.
  • Build the narrowest prototype that tests a risky assumption. Use its behavior to compare design options and record why the team selected one.
  • Plan a staged release with named service owners on the client team. Rehearse fallback and recovery, agree on useful traces, and leave a decision record for future changes.
  • Map the boundary between a recommendation and a committed business action. Identify the authoritative record, who may change it and how the surrounding workflow recovers if a model response arrives twice or arrives late.
  • Help the client choose what to buy, build or defer. Compare the integration effort, operational burden and exit cost of credible options. Record which assumptions would make the team revisit the choice.
  • Define the smallest useful rollout. Specify a group of eligible transactions, a way to observe outcomes and a fallback the service owner can exercise without bringing the whole workflow offline.

Day to day

You’ll use a design document to settle questions that span teams and a working example to test assumptions that a document cannot resolve. For an external dependency, that could mean measuring how a client library behaves when a request times out after the remote service has already accepted it. The result should influence the retry policy, the user message and the record kept for later review.

At an architecture review, give each owner a decision to make. Bring a sequence diagram, an example payload and a small set of failure cases. Capture the decisions in the client’s repository so that whoever changes the integration later has access to the original reasoning.

Projects and first 90 days

A representative assignment

A document service proposes a classification just as a reviewer updates the same record. Map the order in which each system sees the change. Decide where to check the record version, how the interface should show a stale suggestion and which component can safely retry.

A successful handoff includes a working integration test of that race, a sequence diagram and a short recovery procedure. The service owner should be able to exercise the procedure in a rehearsal.

Extend the example into operations. If a reviewer has already acted on an old suggestion, the team needs to know which records were affected. Design a trace that connects the model version and source record to the action a person approved, with access appropriate to each team’s responsibilities.

How your first engagement could unfold

First month. Walk through one production workflow with its owners. Draw it twice, first as a user experiences it and then as a request travels between services. Reconcile the two views and identify the dependency whose failure would create the greatest uncertainty for the person on duty.

Second month. Build a proof of the riskiest integration with the engineers who would maintain it. Test timeouts, stale data and repeated requests. Turn the findings into a decision record, an interface contract and an estimate the client can use to choose a rollout boundary.

Third month. Lead a rehearsal of a limited release. Let each service owner operate their own part and ask an uninvolved engineer to follow the recovery instructions. Use the gaps you observe to refine the design and give the sponsor a clear account of what is ready to take live.

Qualifications

Core experience

  • You’ve taken a multi-service design into operation and can describe a decision you revised after seeing real traffic or operator behavior.
  • You can move between interface contracts, access boundaries and a working prototype, then explain their consequences to an application owner.
  • You’ve worked through competing constraints with multiple engineering teams and recorded a decision they could maintain after handover.
  • You have worked through a production incident or release rehearsal that crossed service boundaries and can explain which technical and ownership changes followed.
  • You can estimate the operational consequences of a design, including latency, capacity, permissions and the time it takes to restore a manual path.

Useful additional experience

  • Experience integrating model serving with a transactional application, event stream, case management system or human review queue.
  • Familiarity with identity propagation, idempotent operations, durable audit records and gradual release patterns.
  • Experience coaching technical leads or working through an architecture decision with a client whose platform and procurement constraints differ from your own.

How to apply

Describe a tricky integration decision. Explain the constraint that mattered most, the alternative you considered and how the service behaved after release.

Choose an example that lets us follow one consequential decision. Explain what the teams disagreed about, how you tested an assumption and what the operator experienced once the design was in use. A redrawn diagram using generic system names is a useful optional work sample.

You can use an anonymized example for confidential work. The application form includes space for your account of the project, an optional résumé and an optional link to a work sample. It currently lets you prepare and review a draft in your browser.

Equal opportunity and accessibility

ENNPRO welcomes applicants of all backgrounds. We consider candidates on the skills and experience relevant to the role, including experience gained through different educational and career paths. If you need an accessible version of these materials or an adjustment to the application process, write to contact@ennpro.pro.

Start application