I’m a computer science student who works like a technical PM: I build my own prototypes and test my thinking against real data.
13,000 Play Store reviews narrowed to 15 complaint themes, scored against 182 reviews I labelled by hand.
ChatGPT voice mode adoption in IndiaFellowship research
36 survey responses. 5 one-to-one interviews. 3 hypotheses tested. 4 solutions ranked by effort and impact.
Starting thesis
- Low usage is mainly a discoverability problem
What the research showed
- Non-users can find it, they prefer typing
- Users go quiet when others are around
- Mixed-language speech is misheard
The thesis I started with, and what 36 survey responses and five interviews pointed to instead.Read the case study
VoC CopilotCuriosity project
13,000 reviews over 30 days. 4,126 with real content. 1,633 inside a theme. 182 labelled by hand.
84% 1 or 5 stars16% 2 to 4 stars
Star ratings across the 13,000 reviews: 10,980 are either 1-star or 5-star.Read the case study
GrowPulseRAG product
12,800 Groww reviews. 49 automated checks with no AI in them. Nothing is published without a typed approval.
Complaint only247
Unrelated42
Confused about a fee28
Praise23
340 reviews matched fee keywords. The model sorted them into four labels; code did the counting.Read the case study