When Digital Divide Data opened its Dehradun centre, it hired 150 people in six weeks, without a single job posting. Many walked in after spotting the signboard. Now, the impact-sourcing pioneer, 25 years old and built to create opportunities for underserved youth, is training Garhwali graduates not to draw bounding boxes but to use, audit, and fine-tune AI. In this conversation with CIO&Leader, COO Steve Larson explains why Dehradun beat Noida and Gurgaon, how freshers become client-ready in weeks, why low-resource languages matter, and how he expects headcount to cross 1,000 within a year.
CIO&Leader: AI and data work—annotation, model evaluation, and LLM support—are increasingly commoditised globally, with margins under pressure. How does an impact-sourcing model compete on cost while preserving the mission and the work?
Steve Larson: LLMs come in different forms, especially across high-resource and low-resource languages. At DDD, we are most passionate about low-resource languages, such as Gawali and other languages with far less data than English or even Hindi.
Hindi is now considered a high-resource language. So parts of our business focus on low-resource languages, which you can still access in relatively low-cost locations. That is where the labour is, because that is where you will find native speakers.
CIO&Leader: DDD has centres across several countries. Why was Dehradun chosen over other Indian tech hubs, and what factors did the site-selection process consider?
Steve Larson: We did look at other locations, including NCR, Noida, Gurgaon, and Lucknow. We chose Dehradun for two reasons. First, it offers very strong young talent.
There are tremendous universities in the area, including WIT (Women’s Institute of Technology), which focuses on the people we want to help most. So the decision was, first and foremost, about where we could have the most impact in the communities where we want to work.
That was a big factor. The costs are different from what we had priced, and a little less costly than Noida. But if you can find pockets within Noida, the costs are not that dramatically different.
CIO&Leader: Walk me through the AI and data work when a client sends model evaluation or LLM support. What does that pipeline look like in Dehradun?
Steve Larson: It depends on where the client is in the model development process. A very early-stage customer needs a lot of data, and data is the most important part of building an AI model. DDD helps with that. If a client has a more mature model, they need evaluations of how it is working. That is called fine-tuning. Some strategies are used to ensure those models improve for each user.
And it is all the steps in between. What we have on the floor today is mostly data ingestion feeding into what is already a very strong, established commercial tool—many of which we probably have on our phones today.
That will then evolve into the actual development of those language models, including testing, evaluation, and fine-tuning.
CIO&Leader: You’re hiring fresh graduates for potential and aptitude rather than experience. What does the in-house training pathway look like, and how long does it take to become client-facing in AI work?
Steve Larson: The training ranges from two to four weeks. We have a two-week injection training that teaches them what AI and ML are and gives them hands-on experience with the tools. We provide that training before we even know which project they might be assigned to.
Then they move on to a client project after about two weeks. After that, they may have one to two weeks of additional client-specific training. Many of our clients have their own tools and applications, so they receive training for them.
Some client training sessions are as short as a week. Some take up to two or three weeks. In the first week, we cover the generic aspects. Then it is customer-specific training. Sometimes we also have customer training.
That is based on their requirements. We get the workforce ready before the customer meets them. And there are two types of people we have.
One is fresh out of college, and the other is already about two years out, with a couple of bachelor’s degrees and no jobs. So they are more mature and grasp things better than the freshers. We match their training needs accordingly.
CIO&Leader: 150 hires in the first six weeks against a 200-seat target is a fast ramp. What’s driving that pace, and what’s the plan once you reach capacity?
Steve Larson: That 150 has been driven with no job postings. It has really been people who just showed up. We had the good fortune of having a partner in Dehradun. Many of those employees have joined our RDDD team.
But the majority of our hires have really been people who just walked in the door. They see our building, our sign, that we are hiring, and they come in. So the pace has been driven mostly by local youth looking for opportunities. Even though I have been here for three weeks, people have come into my cabin to talk about what we do and give me their CVs
So it has been quite amazing. We already have another 35 who have been shortlisted. They will be onboarded, and we are onboarding another 100 in November. They are already lined up, shortlisted, and following up. I should be clear.
I am immensely proud that we are already at 150 associates, or 200 if you include the other staff we have. I did not expect we would be this big this fast. I told Bethul several months ago that if we had 50 people by the end of September, I would be quite proud, and now he has three times that.
To answer your question: what do the next three or four months look like? By the end of this calendar year, I expect we will have around 500 employees, and if you visit a year from now, we will have over 1,000.
CIO&Leader: Impact sourcing has been DDD’s model for 25 years, largely built around outsourcing and BPO-style work. Does AI and data work change who you can recruit, or does it demand a different kind of candidate?
Steve Larson: Not unique to DDD. Many transaction-processing jobs will increasingly be upskilled to use AI to perform those tasks, and the same is true at DDD.
So rather than teaching our employees how to draw a simple bounding box, we are now teaching them how to use AI to complete those tasks, how to perform QA on an AI that might already do that task, and how to fine-tune. Drawing a bounding box seems simple, but on a very complex image with occlusions and hidden objects, it becomes more challenging. So we are training our employees on level 9 and level 10 complexity instead of level 1 and level 2. Full-loop fine-tuning is increasingly important with these models.
CIO&Leader: What does ‘the wider Garhwal and Kumaon regions’ mean in practice? Are you recruiting outside Dehradun city, and how are you doing it?
Steve Larson: The entire 150 is from Dehradun, at least in terms of residence. And 95% are Garhwali, with the rest being Nepalis and Tibetans. You could see the cultural show yesterday. It was full of Garhwali material, right, and everything. The other 5% are Nepalis and Tibetans, who are also kind of Indians here. They have other cards and everything. We also give them opportunities. They come over here.
CIO&Leader: India already has a mature AI/data-services industry with large domestic players. How does DDD’s impact-sourcing angle compete, and how does it differ from companies not built around a social mission?
Steve Larson: We look for a different labour market. We are not looking for employees who live in the big metros and have easy access to those organisations.
We very specifically want to help employees in the local Dehradun community. It is a dream of mine that if an employee could really build skills at DDD, they could find an excellent opportunity at one of those larger organisations. It is not lost on me that that is a dream for many youth.
They want to work for one of those big companies. I do not see DDD as a stepping stone by any means, but I certainly see us as a way to give employees not just great opportunities at DDD, but great opportunities throughout their careers.
CIO&Leader: Twenty-five years in, has DDD’s definition of ‘impact’ evolved? Does an AI annotation job in 2026 create the same economic mobility that a BPO job did in 2001?
Steve Larson: It does, and we measure that in several different ways. We are not focused on basic computer skills. 25 years ago, when we first started this mission, employees had never even opened a laptop before joining DDD. We were teaching them how to type on a keyboard. That is becoming less required. Today’s youth have greater access to these advanced technologies.
Even the youth we celebrated last night, while they may come from underprivileged homes and backgrounds, all have smartphones. Many of them know how to use those things better than I do. It is less about the skills we are teaching now and more about how to apply those skills in a business setting, and how to use them more from a learning agility perspective —less from a learning ability perspective—two very different things.
CIO&Leader: As AI systems mature, some of this annotation and evaluation work risks becoming automatable itself. How does DDD think about the shelf life of these roles, and what’s the path for someone who starts here five years from now?
Steve Larson: The path for somebody who starts five years from now is that many of the employees on the floor today will grow into supervisory and management roles. They will become customer-facing.
Many of our customers today have problems that they themselves do not know how to solve. I expect many of these employees will know how to solve those problems for their customers better than the customers themselves. So they’ll become, within our solutions group, some of them might be studying engineering.
They’ll become part of the teams that actually build these models, rather than the teams that test them or gather the data to build them—and many, many steps in between.