Founding Tech Lead, AI Agents for Back Office (SF Onsite)

ROIAI One; San Francisco (CA); Sur place; USD 180 000 - 220 000; Plein temps

Résumé du poste

ROI-AI, located in San Francisco, seeks a Full-time Engineer to develop AI agents for wholesale distribution's back office operations. You will work alongside the founder and advisor, shaping the architecture from day one. Ideal candidates have strong Python and TypeScript skills, with experience in backend software and integration of external systems.

This role offers a competitive salary of $180K–220K plus equity, emphasizing authority and impact in a small team.

Qualifications

• 5+ years shipping production software, backend-heavy.
• Experience with multi-step agent or tool-use pipelines.
• Strong Python and TypeScript across FastAPI/Node.

Responsabilités

• Build agentic pipelines that work alongside employees.
• Develop eval infrastructure that gates every change.
• Integrate tools for customer systems like ERPs and Postgres.

Connaissances

Backend software development

Python

TypeScript

Integration with external systems

LLM systems understanding

Description du poste

Ship the AI agents that run our customers' back office

Full-time

$180K to $220K + 1% to 2% equity

Start: ASAP

About the Role

ROIAI One is building an AI workforce for wholesale distribution. AI agents that act as digital employees and take on the analyst-level operations work inside a company's back office. They gather documents, coordinate with colleagues, and submit requests through email and retailer portals.

Our wedge is chargeback dispute recovery, a brutal manual back-office problem where we already have paying customers and real revenue. That proves the model. The mission is bigger: deductions, tariffs, AR, compliance, order entry, and beyond. The code you write touches real money the day it ships.

You will be the first engineer alongside the founder (ex-DeepMind, co-author of frontier model research) and our AI/ML advisor, building the platform from the foundation. You own the architecture and the eval bar from day one, not just feature work. Small team, deep work, real authority. The natural path is into technical leadership as we scale. Stack is an agentic pipeline (Roy), with Python and TypeScript.

What You'll Build

• • Agentic pipelines that work alongside employees: Gather documents, coordinate across teams, submit requests through email and portals, and know when to hand back to a human.
• • Eval infrastructure that gates every change: Golden cases, regression suites, scoring rubrics, dashboards the team actually trusts.
• • Tool integration layer for customer systems: ERPs, retailer portals, S3, Postgres exposed as typed tools with audit trails.

Requirements

• • 5+ years shipping production software (or 3+ with clear staff-level ownership), backend-heavy
• • Built and shipped a multi-step agent or tool-use pipeline on a frontier model API, not just single-turn prompting
• • Shipped an LLM system to real users and can describe the eval harness that caught regressions: golden cases, scoring, what you did when a metric dropped
• • Built backend integrations against external or legacy systems with messy or undocumented schemas (ERPs, third-party APIs, SQL Server, file feeds)
• • Owned correctness for an output that mattered (financial, legal, or safety) with evals or tests that gated releases
• • Strong Python and TypeScript across FastAPI/Node, React, and PostgreSQL
• • Onsite in San Francisco, thrives in a small high-ownership team

Nice to Have

• • Workflow orchestration (durable, retryable multi-step jobs)
• • AWS infra ownership (containers, serverless, secrets, S3) and CI/CD
• • Knowledge of EDI, supply chain, or back-office fintech
• • First-5 engineer at a prior early-stage startup

Compensation

$180K to $220K base, 1% to 2% equity, depending on experience (4-year vest, 1-year cliff). Onsite in SF with the founding team. You shape product, architecture, and culture from day one.

Ready to Apply?

Send your resume, GitHub, and one LLM-powered system you have shipped.