[2026]
SC³: The Multi-Solvent Solubility Challenge and Benchmark accepted at NeurIPS 2026. Publication details ↗
THE PERSONAL ARCHIVE
Experience & news.
Experience and news preserved from the original site. Dates reflect the original entries.
News archive
[Nov 25]
[Oct 25]
First-author paper "Panorama: Fast-Track Nearest Neighbors" submitted to arXiv! Developed data-adaptive learned transforms for 2-30× speedup in nearest neighbor search.
[arXiv]
[Jul 25]
Completed summer internship at
RAIL - Carnegie Mellon University
! Worked on machine unlearning, concept forgetting, and neurosymbolic architectures.
[LinkedIn]
[Apr 25]
[May 25]
Received CSE Research Acceleration Fund (RAF) grant for research with
Dr. Sayan Ranu
to present at
ICLR 2025
.
[LinkedIn]
[Feb 25]
Paper "Bonsai: Gradient-free Graph Distillation for Node Classification" accepted at
ICLR 2025
!
[LinkedIn]
[Nov 24]
Spoke at
American Chemical Society
and BioPractify's 38th Students Journal Club on AI for Applied Sciences.
[LinkedIn]
[Dec 24]
Team Pi Propulsion wins
National Champions
at
Lam Research Challenge 2024
- prize of INR 5,00,000.
[LinkedIn]
[Nov 24]
Achieved
4th place nationally
at
Goldman Sachs India Hackathon 2024
among 8,000 participants.
[LinkedIn]
[Sep 24]
[Aug 25]
Received Danish Data Science Academy Scholarship to work with
Dr. Panagiotis Karras
at the
University of Copenhagen
on high-dimensional similarity search.
[LinkedIn]
[Jul 24]
First-author paper published in
ACS Journal of Chemical Theory and Computation
!
[LinkedIn]
Experience
Building
ramAIn
: AI agents that automate any manual UI task across web portals and legacy systems — deployed in minutes, not months. Backed by
Y Combinator
(W26).
Working with
Prof. Akhil Arora
at CLAN, Aarhus University, and
Dr. Lars Klein
(
EPFL
) on adaptive reasoning, multi-agent systems, and post-training of large language models.
Investigating the stability and reliability of LLM reasoning, including the ReasonBENCH benchmark.
Worked with
Prof. Pradeep Ravikumar
in the Statistical & Symbolic Learning Group, MLD.
Research focus on formalizing unlearning, designing mechanisms with provable removal guarantees for data and representations.
Developing NeuroSymbolic architecture for lifting neural models via symbolic modules for interpretability and convergence.
Worked with
Dr. Panagiotis Karras
in Software, Data, People & Society, DIKU.
Developed PANORAMA: a machine learning-driven approach that tackles the ANNS verification bottleneck through data-adaptive learned orthogonal transforms.
Achieved 2-30× end-to-end speedup with no recall loss across diverse datasets from CIFAR-10 to modern embedding spaces including OpenAI's Ada 2 and Large 3.
Working with
Dr. Sayan Ranu
in Data Science and Information Retrieval Lab (DSIRE).
Developed Bonsai: linear-time graph distillation algorithm achieving 22x faster processing across 6 real-world datasets.
Achieved state-of-the-art accuracy in 14/18 test scenarios, outperforming other baselines by ≥5% in multiple cases.
Worked with
Dr. Tarak Karmakar
in Computational Chemistry, Materials & Biology (CCMB).
Developed state-of-the-art GNN framework for solubility prediction using novel Merged-Molecule approach.
Achieved R² of 0.767 and MAE of 0.78 on test set; average MAE of 0.79 across 65 solvents.