Tatva is a sizeable news-reporting Instagram page. The founder, Yashodhar Gulati, was spending real time every day browsing news, picking what to cover, drafting taglines, and trying to predict what would land. The ask was: replicate that judgement loop in a tool.
What it does
- Real-time news access. Continuously ingests current news.
- Tatva’s full posting history. Every post they’ve ever published, indexed.
- Pattern mining. What kinds of stories has Tatva covered before? What framings worked? What level of edge / controversy actually performed versus quietly flopped?
- Live performance data. Hooked into their Instagram analytics via Facebook Business Suite — so the model isn’t guessing what worked, it knows.
The output surface was a chatbot. Yashodhar could ask “what should I post about right now” and get curated trending topics framed in Tatva’s existing voice, or paste in a story he was considering and get a propensity read on whether it would land.
Outcome
Built and working, never fully shipped. As LLMs got better at handling content directly, the value of a custom pipeline narrowed — most news in this category is rephrased and re-posted across channels anyway, and the gap between “ship a custom tool” and “use a foundation model” closed. Project paused, never revived.