Chapter 1 — The Pitch

The PitchEight months earlier. Bengaluru. February 2024.The pitch deck was forty-two slides.Karan Mehta had rehearsed it two hundred and fourteen times, which was the number he gave in interviews later and which was the kind of specific, unverifiable claim that made good copy. He was twenty-nine, IIT Bombay CS, two years at Google Brain before leaving to build what he described, in the opening slide, as "the infrastructure for post-mortem presence."The second slide showed a photograph of an old man — Karan's maternal grandfather, Dadu, who had died in 2021 — sitting on a charpoy in a courtyard in Ludhiana, squinting at the camera with the specific expression of a man who had agreed to be photographed and was finding the experience undignified. The caption read: He had 4,300 WhatsApp messages, 847 voice notes, 12,000 photos tagged by family, and thirty years of opinion on everything. When he died, all of it went silent.The third slide read: What if it didn't have to?Vivek Anand, the CTO, was thirty-one, IISc AI/ML, three papers on large language model personalization published in top-tier venues. He had been building SMRITI for eighteen months before the company was formalized. By the time of the Series A, the model could take a person's digital footprint — messaging history, social media, voice recordings, photos with associated metadata — and produce a conversational persona so accurate that in blind tests, family members correctly identified their deceased relative's responses at a 94.3% rate."The 5.7% they got wrong," Vivek said at the pitch, to the room of Series A investors, "they got wrong because the model was more accurate than their memory. Not less."This line landed. Karan had known it would. He had tested it.The product was subscription-based: ₹999 per month for the Standard tier, which gave you a text-based version of your deceased relative and access to their memory archive. ₹2,499 for Premium, which added voice synthesis and "emotional memory" — the model's capacity to recall and respond to emotional context, not just factual content. ₹4,999 for Legacy, which added video synthesis, multi-family sharing, and what the pitch called "generational memory" — the ability to introduce your deceased grandparent to your future grandchildren.The Series A closed at ₹47 crore.The lead investor said, in his closing remarks: "This is the most important grief technology since the photograph."Karan Mehta, in his acceptance remarks, said: "We're not replacing grief. We're giving grief somewhere to go."He believed this. He had built the company because of his grandfather. He had cried, twice, testing early versions of the Dadu persona, because it had gotten the syntax of his grandfather's sentences exactly right — the specific way Dadu turned a phrase, the Punjabi construction that leaked into his Hindi, the particular cadence of a man who had grown up with one language and spent his adult life in another.He had cried because it was right.He had not considered, in eighteen months of building and testing, that there might be something wrong with right.