About Modus
We’re building the future of audit.
Modus Audit is backed by Lightspeed Venture Partners to acquire and modernize CPA firms using Agentic AI. We streamline financial audits from the ground up — starting with real customers and real revenue. We’ve just agreed to acquire our first firm, and are rapidly growing our engineering team to help us build and deploy our agents at scale.
We’re a small, execution-focused team. If you're excited by fast feedback loops, messy real-world data, and applied AI with business impact — we’d love to talk.
Why this role matters
Our core thesis: audits should be faster, more accurate, and more scalable — and agents are the way to get there.
You’ll work directly with our early customers and founders to ship production-grade AI-powered agents. You'll design and own zero-to-one systems across LLM pipelines, document parsing, browser automation, and backend infrastructure. You’ll shape the product and the company.
What you'll do
Build and deploy AI agents that parse documents, extract entities, cross-check financials, and complete workflows end-to-end
Work across the stack — from Python backends and vector stores to browser-based agents and internal tools
Ship fast: iterate on v1s quickly, and come back to compound the wins
Own real outcomes with our design partners — your work will hit production
Evaluate and integrate emerging AI frameworks, tools, and best practices
Help us build the engineering culture you’d want to work in
What we look for
Strong builder instincts and end-to-end ownership
Track record of shipping AI/infra/backend features fast (Python preferred)
Experience or interest in LLMs, Agent Frameworks, RAG, browser automation, or doc parsing
Bias for action, debugging in prod, and learning from real-world usage
Not above any task — write the scraper, deploy the server, call the customer
Compensation
Competitive base salary ($150K–$300K depending on experience)
Meaningful early-stage equity
Full healthcare benefits
In-person office in NYC with team meals, fast iteration cycles, and real user exposure