Admiral (Senior Full-Stack / AI Engineer)

TL;DR: We're building Admiral, a marketing-intelligence SaaS for DTC e-commerce brands. The MVP is built and running; we're in pilot with our first customer and need a senior engineer to help us ship it to production and keep building. Stack is Next.js 15 + React 19 + Convex + ClickHouse + Anthropic Claude (agentic AI). Hourly, part-time, long-term. About the product Admiral unifies a brand's Meta Ads, Google Ads, and Shopify data into a single analytics layer and puts an AI "briefing" layer on top of it. Instead of logging into five dashboards, an operator gets: - A unified dashboard — revenue, orders, spend, CPO, MER, AOV — with channel tiles and Campaign → Ad Group → Ad drill-down - An attribution / "Sales Lift" view (1st-party revenue vs. pixel attribution vs. unattributed lift) - An automated insight engine that flags meaningful changes and explains likely drivers - Conversational analytics — ask questions in plain English, get data-grounded answers with citations - Scheduled AI briefings and event-triggered analyses (spend-pace, creative fatigue, CPA/ROAS breaches), produced by a multi-agent system (an orchestrator that routes to specialist agents — Meta performance, Meta creative, Google performance, data analyst — and synthesizes one branded response) We're a small, fast-moving team. The product is real and feature-rich; we're now hardening it, deploying it, and expanding it with a long-term engineering partner. Our first pilot is a DTC beverage brand. Where we are today - Built and running locally: auth + multi-org/multi-location, a 16-table Convex schema, three live OAuth integrations (Meta, Google, Shopify) with scheduled sync into ClickHouse, the full dashboard, attribution view, insights feed, streaming AI chat, the automated insight engine, the specialist-agent orchestration + briefings + triggers + per-org cost budgets, notifications, and CSV/XLSX/PDF export. ~13 test files. - Not yet in production. Vercel deploy is planned and documented but not executed. - Known, documented backlog (we keep a wiki of prioritized tasks) — see "What you'll work on." You'll inherit a clean, modern, TypeScript-strict codebase with a project wiki, written plans, and tests — not a mystery box. The stack - Frontend: Next.js 15 (App Router), React 19, Tailwind CSS v4 - Backend/data: Convex (database, queries/mutations/actions, scheduled functions, auth), ClickHouse Cloud (analytics warehouse) - AI: Anthropic Claude via the AI SDK + @convex-dev/agent (persisted threads, streaming, multi-agent orchestration) Integrations: Meta Marketing API, Google Ads API, Shopify Admin - API (OAuth 2.0) - Tooling: Bun, Biome, Vitest, Sentry, Vercel What you'll work on First (paid trial, ~2–3 weeks): get to production & close the top gaps - Execute the first Vercel production deploy (Convex + Next.js atomic build) — the runbook is written - Run the live end-to-end smoke test of AI chat against Anthropic streaming in prod - Make insights respect the active date range (today they're scoped to their generation window) Near-term features - Multiple Meta ad accounts per org from a single OAuth grant (spec'd & written up) - Amazon Ads OAuth (schema is ready; currently manual onboarding only) - Chat hardening — schema-enforced response structure, timeout/retry UX, no-data branch Deeper root-cause grounding for insights (inventory/promo/seasonality signals) Ongoing (the reason this is long-term) - New data integrations and product features as we onboard more brands - AI/agent engineering — extend the orchestrator + specialist agents, prompts, triggers, and insight quality; tune cost/latency - Observability (wire up Sentry properly), test coverage, and general productionization toward multi-tenant SaaS You're a great fit if you - Are a senior full-stack engineer who can own the whole stack end-to-end with minimal hand-holding - Are strong in TypeScript, React, and Next.js (App Router) on modern React (Server Components, React 19) - Have built real things with Convex (strongly preferred) — or are clearly fluent in a comparable reactive/serverless backend (Firebase, Supabase, edge functions) and can ramp fast - Have shipped LLM-powered features in production — the Anthropic / OpenAI SDKs, structured output, streaming, prompt design, and ideally agentic/multi-agent patterns or tool use - Have integrated third-party marketing/commerce APIs with OAuth 2.0 (Meta, Google Ads, Shopify, TikTok, etc.) — token refresh, rate limits, pagination, backfills - Care about data correctness in an analytics product and can reason about SQL/warehouse queries (ClickHouse a plus) - Communicate clearly in writing and work well async Bonus points - ClickHouse or other columnar/OLAP warehouse experience - DTC e-commerce / paid-media / marketing-analytics domain knowledge (MER, ROAS, CPO, attribution) - Vercel + Convex production deployment experience - Bun, Biome, Vitest - An eye for clean, data-dense dashboard UX How we work Async-first, written plans and a project wiki, small PRs, lint/format/type-check in CI. We value engineers who leave the codebase better than they found it and who flag problems early. Expect a short written/async kickoff and a paid trial milestone before we ramp hours. To apply, please include - A short note on a production app you've built with Convex (or the closest equivalent) — what it did and your role. -An example of an LLM/AI feature you shipped to real users — model/provider, and one hard problem you solved (streaming, structured output, agents, cost/latency). - An OAuth integration you built with a major marketing or commerce API (Meta / Google Ads / Shopify / similar). - Your weekly availability, time-zone, and hourly rate. Optional: a link to GitHub / portfolio with relevant work.