Careers
Build the system of record for AI in regulated industries.
Banks, insurers and employers put generative AI into production faster than the controls around it. Meilynx closes that gap: policy enforced inline in the AI request path, and evidence an examiner will accept.
Our founders are former engineers from Google, Microsoft and BlackRock, with decades of building and operating large-scale enterprise systems between them.
Nobody has built this yet
Governing AI inside a regulated firm has no established playbook. The standards are new, the enforcement point is new, and what counts as evidence an examiner will accept is still being settled.
Early stage, real traction
Significant ownership and impact at a company with conviction about where this category goes. Meilynx runs in production inside regulated firms today and holds a SOC 2 Type I report.
Reliability, latency, security, ML
Enforcement sits in the live request path, so the work is held to a latency budget measured in milliseconds, to the security expectations of the firms we run inside, and to model behavior that holds up to review.
Open roles
These are the areas we are actively building. If you are close to a role without matching it exactly, apply and tell us where you land, and if none of them fit, send a general application. We read every one, and we open roles as the work demands them.
Engineering
Staff / Principal Engineer, Data Plane
Bellevue, WA or remote (US) · Full-time
Own the path that enforces policy inline in customer AI traffic and produces the evidence behind every decision.
Full descriptionHide description
Meilynx sits in the AI request path at regulated firms. Policy is applied before a request reaches a model provider and again after the response comes back, each decision is sealed into a tamper-evident audit trail, and all of it runs inside a latency budget our customers hold us to.
You would own that path: its architecture, its performance envelope, and the standard its code is held to. The bar for correctness is high here, because the output is a compliance record a regulator may read years from now, and it has to be right the first time.
This is a senior individual contributor role with real architectural authority. The calls you make are the ones the system lives with for years, and you would be the person other engineers bring a hard design to.
What you would do
- Design and build enforcement in the live request path, including how each control behaves under load and at the edges.
- Own the performance envelope. Overhead added to a customer request is a commitment we make in writing, and it would be yours to hold.
- Build the evidence path: append-only records, integrity verification, and the deterministic behavior an auditor's reproduction depends on.
- Extend model-provider coverage, keeping compatibility faithful through the details that only surface at volume.
- Set the engineering standard around you through design review, testing strategy, and the bar the codebase is held to.
- Carry production alongside the team. The system runs inside customer environments throughout their business day.
Six months in
- You are the owner of the request path, from its design through its production behavior.
- Enforcement you designed is running in customer environments and holding its latency commitment.
- The design review and testing standards the team works to are ones you set.
- Engineers bring you the hard problems before they start building.
What we are looking for
- Staff or principal level experience as an individual contributor, or equivalent scope under a different ladder.
- Deep systems programming experience in Rust, C++, or Go, in a service that carried real production traffic.
- You have run something latency sensitive in production, and can talk through what you measured and what you changed because of it.
- Fluency with async networking: proxies, streaming, backpressure, timeouts, retries, and the failure modes of each.
- A strong instinct for correctness under load. Predictable behavior in the difficult cases is what our customers are buying.
- You write for the person debugging this at two in the morning, which some nights is you.
Nice to have
- Experience with model-provider APIs or LLM serving infrastructure.
- Security or regulated-infrastructure background: audit logging, cryptographic integrity, key handling.
- You have set technical direction across a team without managing it.
Engineering
Senior / Staff Machine Learning Engineer
Bellevue, WA or remote (US) · Full-time
Own the models behind our controls: datasets, training, evaluation, calibration, and a serving path fast enough for a live request.
Full descriptionHide description
Several Meilynx controls come down to a model making a judgment about a piece of text while a request is in flight. That judgment becomes part of a compliance record, which sets a higher bar than most classification work: a miss is a control that did not fire, and a false alarm is a business process that stops.
You would own that layer end to end. Datasets and labeling, model selection and training, the evaluation harness that decides whether a change ships, calibration and thresholds, and the inference path that has to answer inside a strict budget on a live request.
The constraints are the interesting part of this job. Small enough to serve inline, accurate enough to stand behind, and explainable enough that a reviewer who is not an ML engineer can see what fired and why.
What you would do
- Build and maintain models that run inline on live traffic within a strict latency budget.
- Own the evaluation harness: held-out sets, regression suites, and the measured bar a model change clears before it ships.
- Work the precision and recall tradeoff deliberately with the people who live with the result. Setting a threshold here is a policy decision.
- Do the data work properly: labeling guidelines, sparse labels, and distributions that move underneath you.
- Make model behavior reviewable. A control that fires has to be explainable to a compliance officer.
- Ship the serving path with the engineers who own the request path, including what the system does when inference runs long.
Six months in
- A model you built, evaluated and calibrated is running on customer traffic.
- Ship decisions on model changes are made against an evaluation bar you defined.
- The team reaches for your harness before it reaches for a bigger model.
- A reviewer outside engineering can read why a control fired and agree with it.
What we are looking for
- Senior or staff level production ML experience in NLP. You have trained, evaluated and shipped text models other people depended on.
- Real fluency with evaluation. You reach for a held-out set and a calibration plot before you reach for a bigger model.
- Experience fine-tuning and serving transformer-class models under genuine latency and memory constraints.
- Strength in the data work: building datasets, writing labeling guidelines, and reasoning about label noise.
- You can explain a model's behavior to a compliance officer without condescending and without hand-waving.
Nice to have
- Experience with guardrail, content-classification, or information-security detection models.
- Distillation or small-model work where inference cost was the binding constraint.
- You have worked somewhere a model's output was a record someone else was accountable for.
Customer Programs
Technical Program Manager, Customer Programs
Bellevue, WA or remote (US) · Full-time
Run customer deployment programs end to end, from scoping through production rollout, as the person accountable for the outcome.
Full descriptionHide description
Deploying Meilynx at a regulated firm runs as a program. It reaches the teams running AI applications, the security reviewers, the compliance and risk owners who will be accountable for the controls, and procurement. This role holds all of that together and keeps a clear read on where it stands.
You would own customer programs from scoping through production rollout: agreeing what the deployment has to prove, sequencing the work across their side and ours, keeping stakeholders aligned, clearing security review and the vendor questionnaire, and closing the program out against the criteria everyone agreed to.
It is a technical program role. You would understand the product well enough to scope a deployment, follow an architecture discussion, and answer a risk officer's question directly. Writing the enforcement code stays with engineering.
What you would do
- Own customer deployment programs end to end, with a defined scope, schedule, success criteria and a named owner for each workstream.
- Run technical discovery with engineering, security, risk and compliance stakeholders, and turn what you learn into a deployment plan.
- Drive security review, vendor onboarding and procurement to completion, anticipating what each one will ask for.
- Keep a clear, current status on every program, and surface a dependency risk the moment you see it.
- Close the loop into the roadmap. What customers ask for during a deployment is our sharpest product signal, and you would carry it back.
- Build the assets programs run on: rollout plans, runbooks, readouts, and written answers a stakeholder can forward internally.
Six months in
- Programs you run land on their agreed criteria, and everyone involved knows the status without asking.
- Security review and vendor onboarding stop being the long pole, because you have anticipated them.
- Customer stakeholders treat you as the person who knows where the deployment actually stands.
- Product and engineering are acting on signal you brought back from the field.
What we are looking for
- Technical program or engagement management experience delivering software programs into regulated enterprises such as financial services, insurance or healthcare.
- You have owned a cross-functional program end to end and been accountable for the date.
- Enough technical depth to hold your own in an architecture or security conversation, and the judgment to know when to bring an engineer in.
- Command of the regulatory substance, or the appetite to build it quickly: you can hold a conversation about model risk or information-security controls with the people accountable for them.
- You write well. In this market the written follow-up does most of the work.
- Clear and early with status, in both directions.
Nice to have
- You have been through enterprise security review and vendor onboarding from the vendor side.
- Familiarity with AI, model risk or information-security governance in a regulated setting.
- Experience with infrastructure or developer-adjacent products sold to a non-developer buyer.
Meilynx is an equal opportunity employer. Every qualified applicant is considered without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, age, disability, veteran status, or any other characteristic protected by law.
Tell us what you want to work on.
Pick one of the open roles or send a general application. A link to your LinkedIn or résumé works better than an attachment, and every application is read by the hiring team.