DataDive AI
An on-demand AI data scientist on Twitter.
Evidence & scope
Named as Gagan’s side project and described as an on-demand AI data scientist.
Gagan’s ADPList profile
Receipts
An AI interviewer that turns a project conversation into interview flashcards, with supporting quotes kept separate from the written story.
- Problem
- Explaining a project in conversation can be easier than writing a case study. The details that matter in an interview, including personal ownership, decisions and results, can get lost in the retelling.
- Product decision
- Ask for the missing detail, keep the speaker’s evidence, and leave an unknown open rather than completing a more impressive story.
- Approach
- Receipts guides a written or voice interview and organizes the result into case-study sections. Its interview rules track answered questions, allow corrections and preserve exact supporting quotes alongside the synthesized narrative.
Evidence & scope
The prototype separates the written case, supporting quotes and original answer audio. These are different kinds of evidence about the same account.
Implemented workflowThe public prototype is available. A complete browser voice interview and the usefulness of its generated flashcards still need validation; adoption results are not yet reported.
Current scope
HomeLore
A voice-first family-memory prototype that keeps original words, generated narration and unknowns distinct.
- Problem
- Family stories and practical knowledge often live in conversation. Recording them is one step; finding the relevant detail later and knowing what was actually said are separate problems.
- Product decision
- A useful family archive should return to the speaker’s words and make missing information visible.
- Approach
- HomeLore explores recording Hindi and Hinglish conversations, finding details later and returning to old stories. It limits follow-up questions and keeps source quotes, generated narration and unanswered questions separate.
Evidence & scope
The source includes bounded interview follow-ups and separate displays for narration, source material and unknown information. Sample and live-provider modes are explicit.
Implemented prototypeThe prototype is documented, but I have not established how well the full voice flow works or whether families use it and return to it. Public examples should use fictional or consented material.
Current scope
Paper IRL
Turns saved X bookmarks and linked articles into a finite, private edition that can be printed.
- Problem
- Saved reading grows into a queue that competes with feeds, tabs and interruptions. Saving expresses an intention that the reader may never return to.
- Product decision
- Give the reader a finite edition of what they saved, so there is an end to the reading session.
- Approach
- The prototype preserves selected source material and produces private HTML and PDF editions. The early selection follows the saved material rather than inventing a ranking score.
Evidence & scope
The recorded demo produced a PDF from explicit sample data and checked that access was scoped to the owner.
Prototype records, August 2026Whether readers finish more material, and whether paid printing and delivery work, remain unmeasured.
Open hypothesis
AI video editing system
An AI video editor for my podcast and cricket footage.
Research
Research viral cricket shots and videos.
Break down
Break down and identify their editing styles.
Apply
Apply those techniques to my own cricket footage.
Evidence & scope
Used for my podcast as well as cricket footage. The stages above describe the cricket workflow.
Gagan’s scope update