Rohan Sharma
CEO and co-founder of OpenAlmanac
✽
| OpenAlmanac, equivariant deep learning research |
| Location | Cambridge, Massachusetts |
Rohan Sharma is a machine learning researcher and the CEO and co-founder of The Almanac (now OpenAlmanac), an AI-powered knowledge platform launched in October 2025 with the stated goal of building "the next Wikipedia."1 He co-founded the company alongside Kushagra Chitkara and Divit Sheth.2 His research spans geometric deep learning, representation learning, and equivariant neural networks, with positions at MIT, Carnegie Mellon University, and the Vector Institute.4
Sharma attended the Indian Institute of Technology Delhi from 2019 to 2023, earning a Bachelor of Technology in Electrical Engineering with a GPA of 9 out of 10.1
He enrolled at Harvard University in 2023 to pursue a Master of Engineering in Computational Science, completing the program in 2025 with a 3.92 GPA.1 At Harvard, he was a member of the Council of Student Sustainability Leaders, where the office listed his interest as "the theoretical aspects of deep learning and its applications in the field of climate change and sustainability."3 The NSF PG Scholarship Programme, which awarded him a scholarship, described his work as spanning "geometric deep learning, representation learning, and equivariant neural networks."4
From January 2024 to March 2025, Sharma was a machine learning researcher at MIT.1 His work focused on unifying and benchmarking equivariant deep learning, a subfield that encodes geometric symmetries directly into neural network architectures.7 He studied the effect of the choice of O(3) representations -- the mathematical group describing rotations in three-dimensional space -- in the NequIP architecture and developed a new convolutional architecture that he reported increased parameter efficiency by 21% and time efficiency by 32%.1
From May 2022 to January 2024, Sharma was a machine learning research intern at Carnegie Mellon University, supervised by Roger B. Dannenberg.1 He conducted a study of Expectation Networks for predicting melodic sequences, constructing a mathematical framework and proving results on asymptotic optimality.1 He is acknowledged as a collaborator -- alongside Shuqi Dai -- in the Carnegie Mellon master's thesis "A Study of Statistical and Music-Theoretical Melody Prediction" (CMU-CS-22-153) by Huiran Yu, published in December 2022.6
From April 2022 to March 2023, Sharma was a machine learning research intern at the Vector Institute for Artificial Intelligence.1 His research focused on melody extraction from musical scores using weak supervision and on improving MIDI representation, which he reported reduced token lengths by 30%.1
In October 2022, Sharma co-authored the paper "Clarinet: A Music Retrieval System" with Kshitij Alwadhi and Siddhant Sharma, all IIT Delhi students at the time.5 The paper, posted on arXiv, presented a MIDI-based music retrieval system for piano files that introduced a melody extraction algorithm the authors reported improved recall by more than 10%.5 The system implemented two self-designed retrieval algorithms and a modified Mongeau-Sankoff algorithm, achieving overall recall scores above 94%.5 The arXiv metadata states the work was published as a full paper at the DESIRES 2022 conference.1
In the summer of 2024, Sharma interned as an ML engineer at Magnifi, a fintech product under TIFIN, where he reported improving the platform's response accuracy from 65% to 85% and its chart prediction accuracy from 81% to 98.7%.1 From June to September 2025, he was an LLM engineer at Give, another TIFIN product.1
Between his undergraduate and graduate studies, from June to September 2021, Sharma was a data science intern at DronaMaps.1
In October 2025, Sharma co-founded The Almanac in Cambridge, Massachusetts, alongside Kushagra Chitkara and Divit Sheth.12 The platform crawls the public web, reconciles sources, and generates sourced biographies with timelines and maps of professional connections.2 On its Product Hunt launch, Sharma described it as "an AI biography engine designed for depth, accuracy, and connection," with a disambiguation engine to separate individuals who share the same name.2 His Twitter bio reads "Building the people Wikipedia."8
Before The Almanac, Sharma was building a project called Reverie, described on his NSF PG profile as "a generative encyclopedia designed to make research and exploration more seamless and engaging."4 The Almanac has since been renamed OpenAlmanac.1
Sharma and his co-founders are participants in the Canopy Spring 2026 cohort at Founders, Inc..1