Product manager at Dashverse, working on evals and agent skills for Frameo, our AI video studio. Before that, Meesho, St. Stephen's, and two years preparing for the civil services.
works in the greyFrameo turns a script or a brief into a finished video. I work with the team on how we judge that output: a clear definition of a good film or ad, fixed test sets that every change is measured against, and AI judges that are checked against human review before we rely on them.
Frameo's agent chooses a skill for every request. I built a judge that reviews real conversations for two things: whether the agent made the right choice, and where people needed a skill that doesn't exist yet. That work now feeds a broader one: a clear path for a skill to move from draft to review to the shared library, and eventually to retirement.
Auto-dubbing looked like easy maths: take the shows that work, multiply them across languages. I tested dubbed versions against the originals and looked past the cost savings, at whether people kept watching. The quality bar needed work first, so that's what I recommended.
When I joined, the same question got different answers from different teams. I moved our event pipelines onto OpenPanel, rebuilt the reporting layer, and made a rule that every definition gets written in plain words before anyone writes SQL. It's now what the company reports from.
Homepage and product-page experiments for a lot of first-time online buyers. The part I liked most: SMS, the channel everyone had written off, turned out to be the best way to get people to leave reviews.
Eighty-one cells. No hints. I didn't get any either.
Tap a cell, type a digit.
Clashes go red, so you'll know before I do.
In a world painted with shades of grey, find the beauty in the nuanced colors that lie between black and white.
akshara-qc.See the project →Off the clock: a sudoku in pencil, a chess game I'm losing slowly, and a cricket score I check more often than I'd admit.