AI Engineer (Agentic Systems) Remote

About Octopos

Octopos is the operating system for growth for independent grocery and retail stores in the United States. We combine point of sale with tools that grow revenue and profit: rewards and loyalty, AI marketing, predictive inventory, and product cost and revenue analysis.

Our founder ran Lotus Market, a three-location grocery chain, for 10 years before selling it to build Octopos full time in 2019. Every POS he used tried to serve every industry. Octopos is built by grocery operators for grocery operators, and now serves 100+ independent grocery stores, liquor stores, smoke shops and specialty retailers.

We are founder-led and not VC-funded. Our engineering team is based in India.

The role

We are hiring our first dedicated agentic engineer to design and own the agent system inside Octopos. Our goal is to be the AI-first operator for the store: agents that use each store's own data and memory to suggest the right action to the owner, such as what to reorder, what to promote, and where margin is leaking.

In our world, accuracy is the product. An agent that suggests 10 cases of an item that never sells is worse than no agent. You will set the architecture, the eval discipline and the standards, then teach the rest of the team to build on them.

This is a founding role. You report directly to the founder and CEO.

What you'll do

• Design and build the Octopos agent runtime: the loop, tool layer, per-store memory and context management.
• Build tools that give agents safe, read-only, tenant-isolated access to store data in our Laravel and MySQL platform.
• Ship production agents for real merchants, starting with automated purchase-order suggestions and merchant action recommendations.
• Own evals: build regression sets with our domain experts that catch quietly wrong suggestions, and run them on every prompt, model or tool change.
• Add guardrails, fallbacks and monitoring so agents fail safely, and keep cost and latency within budget per store.
• Decide when a simple workflow with one model call beats a full agent, and keep the system as simple as it can be.
• Set up the agent harness and AI-assisted testing practices for our development team.
• Teach: review the team's agent code, write internal guides, and level up our developers so agent knowledge is not held by one person.

What we're looking for

• 5+ years of software engineering, with at least 1 year building LLM agents or agentic workflows used by real users in production. Demos and tutorials do not count.
• A shipped agent you can walk us through end to end, including how it failed in production and what you changed.
• Hands-on experience writing evals and regression tests for LLM systems, not just prompts.
• Deep understanding of agent loops, tool calling, memory, context windows, and graph or multi-agent orchestration, and when not to use them.
• Strong Python or TypeScript, plus comfort integrating a separate agent service with an existing PHP/Laravel and MySQL backend.
• Solid grasp of production concerns: multi-tenant data safety, observability, retries, cost and latency.
• Clear written English and the ability to explain technical trade-offs to a non-technical founder and a junior team.
• A generalist mindset: you engage in team discussions, contribute ideas beyond your own tickets, and care about the merchant outcome.

Nice to have

• Experience with the Claude API, Agent SDK or MCP, or similar tool-use frameworks.
• Retail, POS, inventory forecasting or e-commerce domain experience.
• Experience building agents over structured business data (SQL, time-series sales data) rather than documents only.
• Prior early-stage startup experience, or having been the first AI engineer somewhere.
• Public work: open-source contributions, writing or talks about agent engineering.

Logistics

Type: Full-time

Location: Remote, India

Reports to: Founder and CEO (Arizona, US)

Working hours: At least 4 hours of overlap with Arizona mornings, about 7:30 pm to 11:30 pm IST

Team: Works alongside our India development team and US product manager

Hiring process

1. Application review: your resume plus a short note on the best agent you've shipped.
2. 30-minute intro call with the founder.
3. 60-minute technical deep dive: walk us through a production agent you built, its failures and its evals.
4. Paid 2–3 day trial project: build a reorder-suggestion agent on anonymized store sales data, with an eval set that catches bad suggestions.
5. Trial review and team conversation, then offer.

To apply, visit: https://octopos.notion.site/41437f49b44480b68345d9409122fe19?pvs=105