What has shipped, and when.

Redrob is raising an $18M Series A2.

The round is the third chapter of a company that spent its first six years not raising anything at all, and it buys one specific thing. It is worth being plain about which.

Where the money has come from until now

Redrob was founded in 2018 as an employer-of-record business. Foreign companies hired staff in India through us, we became the legal employer, and every role they opened became a listing only we had. That business was built on customer revenue. There was no institutional capital behind it for six years.

In 2023 Korean institutional investors came in, and we built the hiring agent on top of the supply we already owned. Job Search and Skill Tests launched the same year.

Since then we have shipped four more surfaces and our own model stack. Most of the model work was paid for by somebody else: $2.1M of non-dilutive Korean government funding, through Scale-up TIPS, Blockchain PoC, the Startup Leap Package and Purchase-Conditioned R&D. Our image model placed third worldwide among open-source models, with weights public under Apache 2.0, and the research runs jointly with Seoul National University and Yonsei.

So the product is built. The models are built. The distribution is built, across 400 campuses and eight years of relationships that cannot be bought in a quarter.

What this round is for

A sales organisation.

That is the honest answer and it is a less exciting one than a product roadmap. We have a B2B layer that companies already pay for, and we do not have enough people to sell it. 1,048 companies pay us today. The Indian market our business layer expands into contains 78.6 million registered MSMEs employing 346 million people.

The gap between those two numbers is not a technology problem and no amount of engineering closes it. It is quota-carrying reps, hired against economics we have already watched work.

What is not changing

The cost base. The whole company rests on the fact that answering a question costs us very little, and that is not an accident of scale we are waiting to arrive. It is three years of engineering the inference stack: our own small models, a router that escalates only when it has to, structured retrieval instead of long-form generation, and heavy reuse of the query patterns that repeat.

A company that spent six years profitable does not spend venture money the way a company that never had to. We would rather be judged on that than on the size of the round.

The full detail sits with our investors. What is public is here, on the company page.