AI Software Engineer | Q-Commerce

What you will actually do

• Embed with one or two teams at a time. Sit with Operations, Pharmacy, Growth, or Supermarket for a few weeks. Learn their vocabulary, watch how they actually work, and find the two or three places where software removes real hours of manual effort.
• Ship narrow, working systems fast. A first useful version in days, not a quarter. Prototypes are expected to be rough; they are not expected to be fake.
• Build with LLMs where they earn their place. Retrieval over internal documents and code, text-to-SQL against our data warehouse, agents that draft and route work, evaluation harnesses that tell you whether any of it is actually working. Also the judgment to say when a SQL query and a cron job are the better answer.
• Integrate with what already exists. Our PHP monolith, MySQL, internal service APIs, and the queues between them. You will read a lot of code you did not write.
• Own the whole thing: scoping, building, deploying, instrumenting, and the awkward conversation when a tool is not being adopted. There is no separate team to hand the hard part to.
• Close the loop back into the platform. When the same tool gets asked for a third time, you make the case for it becoming a real product and help the owning pod take it.
• Write things down. Short, honest docs on what you built, what it costs, where it breaks, and what you would not do again.

What we need from you

• Strong, current, hands-on engineering. Python is the working language for most of this. You should be able to build and ship a service end-to-end (API, data access, deployment, logs) without handholding. 3+ years of production experience.
• Practical LLM application experience. You have built something with an LLM API that other people used. You understand retrieval, prompting, tool/function calling, token cost, latency, and why evaluation matters more than the demo. Framework fluency (LangGraph, LangChain, CrewAI, or equivalent) is useful; framework dependence is not.
• Comfort in unfamiliar, messy codebases. You can navigate a large legacy system, find the seam you need, and change it carefully. SQL is second nature to you.
• You talk to non-engineers well. You can run a working session with an ops lead, disagree with a stakeholder productively, and translate a vague complaint into a specification.
• Bias toward shipping and toward honesty about results. You would rather have one tool in daily use than four impressive prototypes, and you will tell us when something is not working.
• Working English and Persian. Persian for the teams you embed with; English for documentation, tooling, and research.

Nice to have

• Experience in e-commerce, quick commerce, logistics, or another operations-heavy domain.
• PHP + Go reading ability, enough to work confidently inside a legacy monolith.
• Vector databases (Qdrant, pgvector) and embedding pipelines in production.
• Self-hosted or open-weight model deployment, and a sense of when it beats an API.
• Consulting, solutions engineering, or founding experience—anything that made you responsible for the outcome rather than the ticket.
• Data engineering exposure: warehouses, dbt, orchestration.

How we work

• Small scope, short cycles. Engagements are measured in weeks. If something has not shown value in a month, we stop and say so.
• Close to the decision. You will present your own work to the people funding it, including engineering leadership.
• Technology is a means. We have no attachment to any particular model, framework, or vendor, and neither should you.