I am Rajan, a builder at Kalmantic Labs operating between Bangalore and San Francisco. I have been working on AI since the computer‑vision days at IIIT‑H, built a startup during the SIFT era, and later held product leadership roles at Intuit and in enterprise software. I now make early‑stage “bamboo shoot pre‑seed” investments via Upekkna and focus on inference economics and agent infrastructure at Kalmantic.
My core thesis is: No AGI without peak tokens per watt per dollar.
I evaluate ideas by the test: You know an answer is good when a child can repeat it back.
I built a startup back when you hand-crafted SIFT vectors, then worked at Intuit and in enterprise software, going "through a few AI winters" before the timing lined up.
I co‑authored the SecLens paper, with MatterSec Lab.
I co‑authored the PeakWeights paper, with Vamshi Ambati.
My principal tools are peakinfer (code on GitHub) and peakweights (a site and research tool).
I frame the Scaling Law as Satya Nadella's successor to Moore's Law.
I use the Jensen Loop for navigating fog rather than adversaries: descend, identify, verify, execute.
I have authored five books on inference economics, agentic organization, operator frameworks, model simplicity, and enterprise agents, and I curate the “AI Papers That Were Right Too Early” collection, which includes 37 foundational papers such as Transformers, RLHF, and Chain‑of‑Thought. I also run side projects like nameornot.com, ultasuno.com, founderchart, ideaeval, and docopic, and I publish articles on my Substack covering topics from Anthropic competition to first‑principle thinking.