SUMMEDD INC.

Compensation

Full Description

Senior Full-Stack Engineer - AI Agent Systems TaxFlow · Accounting & Taxation platform · Full-time · Remote (USA-based) About TaxFlow TaxFlow is a tax-preparation platform for CPA firms. We take the most document-heavy, deadline-driven workflow in professional services - the annual tax return - and rebuild it around AI: documents arrive, get classified and extracted by an agent pipeline, land in a per-return workspace ("binder") as reviewable, cited values, and flow through preparation workpapers to the firm’s tax software. Firms keep the judgment; the platform does the shuffling. This is real-money, real-compliance software. Taxpayer PII, IRS Pub 4557 / FTC Safeguards obligations, audit trails, immutable records. Accuracy is not a nice-to-have - it is the product. The role You’ll be the second engineer, working directly with the founding team across the whole platform - and you’ll do it in an agent-native workflow: we build with parallel AI coding agents, where engineers set direction, write sharp specs, review rigorously, and orchestrate agent teams rather than typing every line. If your instinct on a complex migration is "spec it precisely, fan it out, verify adversarially," you’ll feel at home. Your center of gravity: • AI / agent systems (the core of the role). Design and harden the document-intelligence pipeline: multi-step agent harnesses for classification, extraction, and review of tax documents (W-2s, K-1s, 1099s, organizers, prior-year returns); prompt/tool design; structured outputs; confidence scoring and citation of extracted values; evals and regression suites so quality is measured, not vibes. • Backend. Postgres data modeling in a strictly multi-tenant schema (70+ tables, DB-first with Drizzle), server functions, background job orchestration (Inngest), document storage (R2/S3), and vendor integrations - CCH Axcess import/export, IRS MeF, third-party extraction APIs - kept behind clean adapter seams. • Full-stack product. Ship vertical slices end to end in our TanStack Start + React app: extraction-review UIs, spreadsheet-grade workpapers, pipeline/kanban views. You don’t need to be a designer; you do need to care that a preparer can trust what’s on screen. • DevOps (a solid working share). Own our path from "works locally" to boring, observable production on AWS: deploy pipelines, CI, environments and secrets, monitoring/alerting, backup and retention posture, and the data-residency/security requirements that come with taxpayer data. Nothing exotic - it just has to be right, because tax season doesn’t reschedule. Our stack (what you’d actually touch) • Monorepo: Turborepo + pnpm, Node 24, TypeScript end to end, Vitest • App: TanStack Start (SSR React), Tailwind v4, shadcn/ui on Base UI primitives • Data: PostgreSQL (Neon in prod, Docker locally), Drizzle ORM, DB-first schema, firm_id tenancy on every domain table • AI: multi-provider LLM stack (Anthropic Claude today, provider-agnostic by design - OpenAI, Gemini, and open models where they fit) driving the agent pipeline for document classification, extraction, and review; structured outputs, per-document extraction specs, eval harnesses • Jobs & storage: Inngest, Cloudflare R2 / S3-compatible storage (MinIO locally) • Auth: self-hosted Better Auth • Integrations: CCH Axcess bridge, IRS MeF, SUMMEDD extraction API - isolated adapters • Hosting: AWS (production); GitHub-based flow What you’ve done before (requirements) • 6+ years building production software, with at least 2 recent years shipping LLM/agent-powered features to real users - agent loops or harnesses, tool use / function calling, RAG or document pipelines, eval frameworks. You can talk concretely about failure modes you hit and how you measured your way out. • Strong backend fundamentals: relational data modeling, transactions, migrations, queues/background jobs, idempotency, API design. Comfortable owning a Postgres schema that other things depend on. • Fluent TypeScript across the stack and at ease in a modern React codebase (we use TanStack Start; Next.js experience translates fine). • Enough DevOps to own production: AWS experience - CI/CD, environment/secret management, networking basics, monitoring, and incident hygiene. You’ve stood up and run production infrastructure yourself, not just consumed someone else’s. • Rigor. You write specs before code when it matters, test what can silently be wrong, and treat "the model said so" as a hypothesis, not an answer. • Comfortable working with AI coding agents as a force multiplier - directing, reviewing, and correcting them - not threatened by them, not blindly trusting them. Nice to have • Fintech, accounting, tax, or other regulated/compliance-heavy domain experience (PII handling, audit trails, retention/WORM, SOC 2 groundwork) • Document-AI specifics: OCR/vision-model extraction, layout parsing, confidence calibration, human-in-the-loop review UX • Experience with any of: Drizzle, Inngest, Better Auth, Neon, Cloudflare R2, Turborepo • Prior early-stage experience - you’ve been engineer #1–5 somewhere and know what "owning it" means How we work • Small, senior, direct. You work with the founding team daily; decisions are made in hours, not sprints. • Agent-native development. Plans and specs live in the repo; AI coding agents do much of the mechanical work in parallel; humans own architecture, correctness, and review. The bottleneck we care about is ambiguity, not typing speed. • Vertical slices, verified end to end. No layer-by-layer phases; everything ships behind quality gates (typecheck, tests) with real data flowing. • Correctness culture. Tax numbers, tenancy boundaries, and audit trails get the paranoid treatment: root-cause fixes over patches, tests over promises. Details • Compensation: USD 120,000 – 130,000 • Location: Remote, USA-based. Mid to West Coast preferred for timezone. You will need timezone overlap with both New York and Sydney, Australia. Rare travel to New York for an in-person meeting may be required. • Start: As soon as you can Hiring process 1. Intro call (30 min) - mutual fit, your AI/agent war stories 2. Technical deep-dive (60 min) - walk us through a system you built: an agent pipeline or a backend you owned; we’ll dig into the decisions 3. Working test session - a realistic slice of our world (e.g. design an extraction eval, or spec + build a small vertical slice with agent tooling) 4. Final discussion and offer To apply Email jobs@summedd.com. Include something you’ve built with LLMs that you’re proud of - a repo, a demo, or a write-up of the hard parts. Pay: $120,000.00 - $130,000.00 per year Work Location: Remote