Technical Delivery Lead, AI Dev Pipeline

Read this part first, because it disqualifies most applicants You will write little or no production code in this role. AttuneAI ships software through a multi-agent AI development pipeline. AI agents do the implementation, the QA pass, the screenshot diffs against Figma, and the pull request write-ups. They are good at it. They are also literal, occasionally overconfident, and completely dependent on the quality of the instruction they are handed. Your job is to be the person who hands them that instruction, watches what comes back, and decides whether it is actually true. If you are looking for a role where you spend your day in an editor, this is the wrong post and we will both waste an interview. If the idea of directing a fleet of AI engineers, owning the infrastructure they deploy to, and being personally accountable for what reaches users is the most interesting thing you have read this month, keep going. About AttuneAI AttuneAI is an AI relationship wellness product. Couples and individuals work through four guided modes (Reflect, Repair, Resolve, Refine) with an AI facilitator. Real-time three-way voice, where the AI facilitates a live conversation between two partners, is on the near roadmap. This matters for the role in one specific way: the product sits next to mental health. There are crisis-safety guardrails, refusal thresholds, consent tiers, and data retention rules in the codebase that are clinical and legal decisions, not engineering preferences. Part of your job is knowing which changes are yours to wave through and which ones stop dead until a human with the authority to decide has decided. We will teach you where those lines are. We need you to respect them without being reminded. We are small, fast, and shipping. Nobody here is going to build you a runway of process. What you will actually do Steer the pipeline. Our pipeline portal has a chat intake (Issue Command) with Plan mode and Build mode, a live status board, and an approvals queue. You will live in it. You will write the task briefs that agents build from, decide the scope, attach the right repo, branch, Jira ticket and Figma frame, push back in Plan mode when the plan is wrong, and flip to Build only when the task is genuinely ready. A vague brief produces a confident, well tested, beautifully documented implementation of the wrong thing. Preventing that is most of the job. Own the approval gate. Every merge passes a human approval step. Today the median wait at that gate is around three and a half hours and the worst case has been seventy two. That latency is the single largest drag on our throughput, and it exists because there has not been one person whose job it is to sit at the gate. That person is you. You will read the diff summary, check the merge readiness, confirm CI is green on the exact commit, confirm the approval still binds to that commit, and either approve, request changes, or escalate. You will also recognise the changes you must never approve alone: schema migrations, pricing and entitlement changes, auth and signing changes, and anything touching safety, consent or retention. Own the repositories. Four active repos across an Angular and Capacitor mobile and web client, a .NET Web API, a Python FastAPI AI service, and the Node pipeline portal itself. Branch protection, required checks, branch hygiene, stale branch cleanup, release branches, GitHub Actions workflow health, GitHub App permissions and key rotation. You do not have to write the app code. You do have to know why a build failed and who to route it to. Own the Azure footprint. Azure, Canada Central, resource groups split by tier (frontend, backend, AI, shared, voice). Static Web App, App Service, Container Apps, Application Insights, Key Vault. There is no infrastructure as code today, every resource is hand managed, and deploy pipelines only update what already exists. You will be the person who knows what is actually deployed, what it costs, what is configured where, and who keeps that documented instead of tribal. Introducing IaC properly is an open project you could own. Keep Jira and Confluence honest. Jira is a machine readable premise for the pipeline, not decoration. If a ticket says In Progress and the PR merged four days ago, agents make bad decisions downstream. You will keep the board true, groom the backlog, write tickets that an AI can build from without a follow-up question, and keep the Confluence architecture pages current when reality moves. Communicate. Daily, in writing, unprompted. What shipped, what is stuck and why, what you need a decision on, what you are worried about. Short, precise, no filler, no burying the important sentence in paragraph four. Catch the things nobody asked you to catch. Our honest list of known gaps includes migrations that auto-apply on deploy with no pre-flight check, agent sessions that can silently die and strand a job, and a portal event log that is a status board rather than a system of record. We wrote those down ourselves. We would like the next one to be found by you. The stack you will steer (not write) Web and mobile client: Angular, Capacitor (iOS and Android), TestFlight, Play Console Backend API: .NET Web API, EF Core, onion architecture, JWT and claims-based auth AI service: Python 3, FastAPI, SQLAlchemy async, Alembic Real-time voice: LiveKit, Deepgram, Pipecat (spike stage, moving to production) Cloud: Azure (Static Web Apps, App Service, Container Apps, Key Vault, App Insights) Source and CI: GitHub, GitHub Actions, GitHub Apps, branch protection Tracking and docs: Jira, Confluence Design: Figma (specs, node inspection, frame exports for visual diffing) Delivery: Our multi-agent dev pipeline portal You need to be literate in all of it. You need to be expert in none of it. The one place literacy is not enough is Azure and GitHub, where you are the owner. What we need from you Non-negotiable: Pacific Time. You live and work in PT/PDT, or you are permanently relocated into it. This is not a "significant overlap" role. We need you in the same working day, reachable, in real time. Publication quality written English. Your Slack messages, tickets, PR summaries and agent briefs are the product of this job. Ambiguity in your writing becomes a defect in the codebase. Your application will be read as a writing sample. There is no separate writing test, because the application is the test. Outstanding attention to detail. Concretely: you notice that a green CI check ran against a stale tree. You notice the ticket says one thing and the Figma frame says another. You notice the number in the summary does not match the number in the table. This is the single highest weighted trait in our evaluation. Self-starting. You will be given an outcome and a boundary, not a task list. If you need a daily standup to know what to do next, you will struggle here. Judgement about scope. We want you to do more than you are asked, inside the boundary. We do not want you approving a migration on a Sunday because it looked fine. Knowing the difference is the job. Comfort being wrong out loud. When you find out you were wrong, we want to hear it from you, immediately, in plain language, before it costs anything. We have written incident reports on our own agents doing exactly this well, and on them doing it badly. It is cultural here. Full time, with genuine availability. Releases, store submissions and voice infrastructure work create real evening and weekend spikes. We will not pretend otherwise. If you have capacity beyond full time, we want it and we will pay for it. Strongly preferred: You have shipped a real mobile app to the App Store or Play Store and know what actually goes wrong: provisioning profiles, signing keys, store review rejections, staged rollouts, in-app purchase configuration. Hands-on Azure. Resource groups, App Service, Container Apps, Key Vault, reading Application Insights when something is on fire. You have run a release process, a QA gate, or a deployment pipeline where you were the last person before users. You have used AI coding tools seriously and can articulate specifically where they fail. Figma fluency at the level of reading a spec and telling an implementer exactly what is wrong with their build. Any working knowledge of real-time media (WebRTC, WebSockets, LiveKit, streaming servers) or speech infrastructure (Deepgram, TTS and STT pipelines). Genuinely a bonus, not expected. Not required: Senior engineering ability. You do not need to be able to write the Angular component or the EF Core migration. You need to be able to read them, ask the right question about them, and know when the answer is unsatisfying. A computer science degree. What will make you fail here We would rather tell you now. Waiting to be asked. Reporting that something is done when what you mean is that an agent told you it was done. Approving a change because it looked fine, rather than because you checked the specific thing that could make it not fine. Treating "the tests passed" as proof. A green suite is a floor, not a proof, and we will expect you to internalise that sentence. Long, hedged, low-information updates. Needing the work broken down for you. Going quiet when something is going wrong. Logistics Location: Remote, Pacific Time required. Hours: Full time. Additional paid hours available and welcome if you have the capacity. Engagement: [employment type and compensation to be inserted before posting] Start: As soon as we find the right person. Access: This role holds production cloud and source control credentials from an early stage. Our vetting is thorough and it is described honestly in the application steps below. Please do not apply if you are not prepared to be verified. How to apply Applications that skip any of the six items below will not be read. This is deliberate. Following a precise multi-part instruction exactly is the core competency of this role, and the application is the first test of it. Apply for this job with a cover note, maximum 400 words. Tell us about a time you were responsible for the last check before something reached users, and you caught something. What was the signal you noticed, what did you do, and what would have happened if you had not. One story, told well, beats a summary of your career. 2. Mobile apps you have shipped. For each app you want to claim, give us all of the following: App name and a live App Store or Play Store link. The dates you worked on it and whether it was employment, contract or personal. Your exact role, stated precisely. Not "worked on the iOS app". We want: which parts were yours, what you personally decided, what you personally shipped, and what you did not touch. One specific thing that went wrong on that project and what you did about it. A named person who can confirm your account of it. If you have never shipped a mobile app, say so plainly in one line rather than stretching something into the shape of one. It is preferred, not required, and an honest gap costs you far less than an inflated claim, which is disqualifying. 3. Three professional references. Name, current role, company, email and mobile number, your working relationship, and the dates. At least two must be someone you reported to or who directly oversaw your work. At least one must be from the last two years. Please confirm in writing that we may contact all three, and please tell them we will be calling. 4. Your answer to this, in no more than 250 words. An AI agent reports that it has finished a task, all tests pass, and it has opened a pull request. The pull request includes a database migration. The agent's summary is confident and well written. You have thirty minutes before the founder asks whether it can be merged. What do you actually do in those thirty minutes, in order, and what would make you say no? We are not testing whether you know our stack. We are testing how you think when something looks fine. 5. Confirm your time zone and where you are physically located.