Job role: Software Engineer (AI-Accelerated)
Duration: Full Time
Location: Phoenix, AZ 85071 onsite
What you'll do:
You'll build and maintain department workspaces in the Console and AR1. Real examples from the last 90 days:
- Build department workspaces end to end. Ticket hubs, queues, tiles and filters for Client Services, Legal, HR and Compliance (e.g. a ClickUp-style filter builder, a performance-review system, a legal case tracker).
- Work on regulated workflows. Credit-dispute (e-OSCAR ACDV) queues with locking, auto-submit and audit trails, where a bug means a compliance issue, not just a broken screen.
- Replace CollectOne, one department at a time. Read legacy SQL Server data, mirror it into Postgres, and rebuild the daily screens staff rely on.
- Write database migrations safely. Plan apply-before-merge order, row-level security, and one-transaction-per-page-load rules.
- Wire integrations. Microsoft Graph mailboxes into tickets, Amazon Connect call screens, AWS Bedrock AI features (reply suggestions, PDF invoice analysis), S3 uploads, Stripe and QuickBooks.
- Drive AI agents well. Write clear prompts and specs, break work into small stacked PRs, and review every line the agent writes.
- Keep CI honest. Respect the guard checks (page index, lint ratchet, migration hygiene, secret scanning) and add new guards when a mistake could repeat.
- Sit with users. Turn a messy request from a supervisor into a clear spec, then show them the working screen.
How we work:
AI writes most of the code; humans own every decision and every merge. The rule underneath it all: if a safeguard depends on someone remembering it, it's already broken, so we make it fail loudly instead.
- Prove the data is live before building on it. Run one real query and show the output.
- Stop, don't guess. If a decision, credential or value is missing, write it down as a blocker and ask.
- Small PRs, opened as drafts. Nobody merges their own work; our lead engineer reviews.
- Grade what you promise. Each change is enforced (a test catches it), code (built, unguarded) or process (a human must remember). Name the process ones out loud.
- PR description as a decision log. What was decided, by whom, on what date, plus a "Not in this PR" line.
- No unmeasured numbers. Never say "tests pass" or "migration applied" without the command that proves it.
- Post-mortem in five lines when something breaks in production.
Stack: TypeScript, Next.js 16 / React 19, Postgres with Drizzle ORM, SQL Server (mssql), Tailwind, Vitest, Vercel, AWS (Bedrock, S3, SES, SQS, Connect), Sentry, GitHub Actions, Microsoft Entra / Graph.
Required qualifications:
- 4+ years building and running production web applications (full stack).
- Strong TypeScript and React; Next.js App Router or similar server-rendered framework.
- Real SQL skill: schema design, migrations, indexes, transactions, constraints. Postgres preferred.
- Daily, hands-on use of AI coding agents (Claude Code, Cursor, Copilot agent mode or similar) to ship real work, with examples you can show.
- The habit of reading and challenging AI-written code: you can explain why the agent's first answer was wrong.
- Writes tests that matter (Vitest/Jest or similar) and knows which risks deserve them.
- Comfortable with Git, pull requests, code review and CI pipelines.
- Clear plain-English communication with non-technical users and managers.
- Judgment about data that matters: PII, financial records, audit trails.
Preferred qualifications:
- Collections, credit reporting (FCRA, Metro 2, e-OSCAR), fintech or other regulated industries.
- Legacy migration work, especially SQL Server to Postgres.
- AWS (Bedrock, S3, SES, Connect), Microsoft Graph / Entra ID, Vercel.
- Payments and accounting APIs: Stripe, QuickBooks Online.
- Built LLM features into a product (prompting, extraction, guardrails).
- Set up or ran autonomous agent workflows with review gates.
- SOC 2 or security-review experience (row-level security, secret scanning, OAuth token handling).