Building Mio, an AI coach that backs every rating with evidence

Most training judges a cricket match by its first and last over. Mio was built to watch every ball: an AI interviewer that checks what engineers actually understand, at every step of a program.

Role

Senior Delivery and Product Lead

Where

Trumio

When

2024 – Present

Worked with

Founders, product team and engineers

• 500–1,000enterprise learners interviewed
• 1,500data points per learner
• 25–30page evidence report per learner
• 100%completion across concurrent enterprise programs

How I got here

I joined Trumio in April 2024. The founder reached out to me, and the founding team had led engineering and business at companies like Microsoft, Cisco and IBM. Trumio already had a product manager, so they offered me the other side: everything that happens after a sale. I took it on purpose. I knew how to build products, but not how to sell them, and here I could learn from a co-founder with 25 years in IT sales.

Trumio's first product was a marketplace where companies posted small projects and students bid to do them. I owned the student side. Instead of emailing college directors, I went through student leaders: Gymkhana presidents, general secretaries and tech club heads. I spoke with more than 5,000 students, onboarded about 2,000, and signed agreements with seven institutions, including IIT Kharagpur, IIT Kanpur, IIT BHU and IIT Bhubaneswar.

Then Trumio pivoted. Companies didn't just need cheap work done. They needed proof that their engineers could do the work, especially now that AI writes so much of the code. That became Trumio Enterprise, a platform where engineers learn by doing real work and show what they can do.

The problem: judging the match by two overs

Most training measures people twice: a test at the start and a test at the end. A coding test tells you where someone is today. It says little about whether they can deliver on a real project, and nothing about what happened in between. It's like judging a whole cricket match by the first over and the last.

Trumio's early programs tried to fix that with people. Technical Project Advisors interviewed every learner and assessed their work. It worked for 5 to 10 people. It couldn't work for hundreds: advisors were expensive, interviews took time, and the analysis piled up. I was the one watching it break, and that was the starting point for Mio.

What Mio does

Mio is an AI interviewer and coach that runs through a whole program, like a Test match where every ball counts.

1. A gap analysis at the start. Mio interviews each learner on the exact tech stack the program targets, tied to a real outcome. Not "learn Java," but "solve Java problems on your project independently and be interview-ready."
2. A path that fits the person. Based on that interview, each learner starts where they actually are, from complete beginner to partly experienced.
3. An interview after every module. Programs have 10 to 20 modules of lessons, videos and coding. After each one, Mio interviews the learner, checks what they learned and coaches them on the spot. It knows the full context: their gap analysis, past projects, résumé and the whole program.
4. Checks inside the code editor. While learners build, Mio pauses them with short questions about code they just wrote or accepted from AI, so understanding is tested in the moment, not only afterwards.
5. An exit interview and an AI code reviewer. Coding work is checked against the task brief, and everything feeds one detailed report per learner.

21@GetMapping("/orders/{id}")22public ResponseEntity<Order> getOrder(@PathVariable Long id) {23    return orderRepo.findById(id)24        .map(ResponseEntity::ok)                          Accepted from AI25        .orElse(ResponseEntity.notFound().build());26}

My part

Mio was a team effort. Here is the honest split.

I was the client-facing person and spoke with participants every day. That made me the bridge: I brought what customers and learners told us back to the people building the product.

What went wrong, and what we changed

The finding that mattered most

In one pilot, engineers scored high on the quality of their code submissions while accepting AI-written code 90 to 100% of the time without checking it. Their work looked good. Mio's interviews showed they didn't understand it.

It also works the other way. Some engineers accepted AI code without checking it, yet explained it well when Mio asked. That is a different problem: a habit, not a gap in knowledge. One needs teaching; the other needs a review step before accepting AI code. Looking only at the output can't tell them apart. Asking people about their own work can.

Results

• Every participant was interviewed at every stage, with no extra headcount.
• Time and cost that used to go to human-led interviews were freed up.
• Data for analysis arrived quickly and in a consistent format, about 1,500 data points per learner.
• The reports helped clients decide which engineers were ready to deploy onto projects.

What I'd take from it

The best ideas for Mio came from talking to learners and clients every day. That's still how I build.