Retailers have started selling their suppliers a weekly account of what sold, where, at what price, and what went wrong getting it to the shelf — down to the item and the store. It is the most detailed picture of retail that has ever existed. Most brands have one analyst and no way to read it, so decisions get made on a fraction of it.
We are building DuitLabs Cortex, which reads the whole thing and tells a brand team what changed, why, and what to do next. We started with Walmart, the largest and hardest of these feeds. Every other retailer follows.
Our founders built Walmart's global forecasting and replenishment systems. 15+ patents. Brands on the platform today run books worth billions.
The work
- Ingest retailer feeds over Delta Sharing. Billions of item × store × week into sub-second reads. The volume is the easy part; keeping it correct and current as feeds change shape is not.
- Own the semantic layer — the retail math the rest of the product depends on. Rolling windows, share, velocity, and telling lost shelf space apart from a real drop in demand. Built once, so the second retailer costs a fraction of the first.
- Build the recommendation layer. MCP servers, LLM pipelines, and evals that catch a wrong answer before a buyer sees it.
- Build the app brand teams open on Monday morning. The output only counts if someone acts on it.
The hard problem here is not scale, and it is not prompting. It is being right. A recommendation that sends a brand team into a buyer meeting with the wrong read costs them credibility they can't buy back. Everything we build is judged on that.
Stack
You
- 4–8 years shipping production code.
- Strong in data engineering, backend, or frontend — and willing to work across all three.
- You've run LLMs in production, not just in a demo. You know what breaks and how you caught it.
- You'll join customer calls. Talking to brand teams is part of the job, not a distraction from it.
- Nobody writes your tickets.
You don't need retail or CPG background. The domain is learnable and we'll teach it. What doesn't transfer is judgment about correctness under ambiguity, which is most of this job.
Working here
Remote from Bangalore. Small team, no layers, and a real overlap window with the US team — we'd rather agree on a few fixed hours than pretend time zones don't exist.
Compensation is competitive and above industry standards for senior engineers in Bangalore, set on experience. When we talk numbers, we talk cash, not CTC.
Apply
Send us something you built and a line on why this problem interests you. Not a cover letter — a link and a paragraph is plenty. We read everything and we reply either way.
[email protected]