About
I'm Founding ML Engineer at Kipo AI. I completed my MS in Scientific Computing at University of Pennsylvania (4.0 GPA), working with Andrea Liu on learning in physical systems and Mark Liberman at the LDC on NLP. I also interned at Cisco Research with Jayanth Srinivasa.
Previously: At Microsoft Research India, I worked on Project Froid (scalar UDF inlining for MSSQL) with Karthik Ramachandra. At RIKEN Japan, I worked on open quantum systems with Franco Nori and Neill Lambert, co-developing QuTiP-BoFiN. At J.P. Morgan, I was a Quantitative Researcher on the Rates desk, building pricing and risk models. At tensoralpha (crypto hedge fund), I worked on crypto derivatives trading strategies with Thomas Schmelzer. I hold a BS in Computer Science from BITS Pilani.
I work on understanding how neural networks learn and reason: geometry of memory in RNNs, multi-step reasoning in LLMs, learning in physical systems.
News
Jan 2025Won 1st place at Exa AI Discovery Agents Hackathon (50+ teams)
Jan 2025Paper on Alignment Trilemma accepted to ICLR 2025 Workshop on Bidirectional Human-AI Alignment
Dec 2024Paper on Neural Networks for Option Pricing accepted to ICLR 2025 I Can't Believe It's Not Better Workshop
Nov 2024Won 1st place at Sentient Foundation Werewolf AGIThon (AGI House)
Oct 2024Paper on Persistent Homology for Tactile Recognition accepted as Oral at NeurIPS 2024 Workshop on Touch Processing
Oct 2024Survey on Foundation Models & Continual Learning accepted to NeurIPS 2024 Workshop on Scalable Continual Learning
May 2024Submitted MS thesis on learning in physical systems with Andrea Liu and Nachi Stern
May 2024Paper on LLM Reasoning via Classification accepted to LREC-COLING 2024
May 2024Joined Kipo AI as Founding ML Engineer (Employee #2)
May 2023Started internship at Cisco Research
Mar 2023QuTiP-BoFiN published in Physical Review Research
Jan 2023Joined Prof. Andrea Liu's group at UPenn Physics
Sep 2022Joined Linguistic Data Consortium with Prof. Mark Liberman
Aug 2022Started MS in Scientific Computing at UPenn
Aug 2021Joined J.P. Morgan as Quantitative Researcher (Rates)
Feb 2021Joined Humit as Founding Backend Engineer
2020CS50 Puzzle Day (Harvard x Facebook): Global winner (8/8)
2020Joined tensoralpha GmbH as Quantitative Developer Intern
2020Participated in NTT Coherent Ising Machine group's research
Oct 2019Joined RIKEN Japan with Prof. Franco Nori
Jul 2019Started internship at Microsoft Research India
May 2019Submitted undergraduate thesis on DMRG for Potts Models with Prof. Sreejith GJ (IISER Pune)
2019Invited to Qiskit Asia Camp by IBM
2019WorldQuant Challenge: Gold ranking for high Sharpe alphas
Jan 2019Started undergraduate thesis research at IISER Pune
Dec 2018Offered WorldQuant WebSim consultant role for exceptional performance
May 2018Selected for ACM Summer School in Cryptology at ISI Calcutta
May 2017Started R&D internship at Zero Labs (NLP)
Publications
QuTiP-BoFiN: A Bosonic and Fermionic Numerical Hierarchical-Equations-of-Motion Library
Physical Review Research 5, 013181 (2023)
Understanding and Controlling the Geometry of Memory Organization in RNNs
Under Review
Enhancing Large Language Models through Transforming Reasoning Problems into Classification Tasks
LREC-COLING 2024
The Alignment Trilemma: A Theoretical Perspective on Recursive Misalignment and Human-AI Adaptation Dynamics
ICLR 2025 Workshop on Bidirectional Human-AI Alignment
On the Limitations of Neural Networks for Option Pricing: Analysis of Volatility Regime Sensitivity
ICLR 2025: I Can't Believe It's Not Better
DAGnosis: A Benchmark for Complex DAG-Based Multi-Step Reasoning in Large Language Models
Under Review
An Initial Exploration of Using Persistent Homology for Noise-Resilient Tactile Object Recognition
NeurIPS 2024 Workshop on Touch Processing (Oral)
Foundation Models Meet Continual Learning: Recent Advances, Challenges, and Future Directions
NeurIPS 2024 Workshop on Scalable Continual Learning
Theses
Learning in Physical Systems
Advisor: Prof. Andrea Liu
Density Matrix Renormalization Groups for Potts' Model
Advisor: Prof. Sreejith GJ (IISER Pune)
Patents
Automatic Retrieval Augmented Generation with Expanding Context
A. Deepak, W. Healy, T. Raheja, J. Srinivasa, R. Kompella, R. Sinha. US Patent App. 18/442,996 (2025)
Cache-Generated Frequently Asked Questions Page
T. Raheja, R. Sinha, W. Healy, J. Srinivasa, A. Deepak, R. Kompella. US Patent App. 18/439,888 (2025)
Memory-Based Function Calling for Logical Formulation and Inference for LLMs
T. Raheja, A. Deepak, W. Healy, J. Srinivasa, G. Liu, R. Kompella. US Patent App. 18/439,966 (2025)
Performance Monitoring, Mitigation, and Retraining of Large Language Models
A. Deepak, R. Sinha, W. Healy, T. Raheja, J. Srinivasa, R. Kompella. US Patent App. 18/440,710 (2025)
Experience
Kipo AI2024-
Founding ML Engineer
Cisco Research2023
ML Research Intern
UPenn - Liu Group2023-24
Graduate Research Assistant (Prof. Andrea Liu)
Linguistic Data Consortium2022-24
NLP Research Projects Programmer (Prof. Mark Liberman)
J.P. Morgan2021-22
Quantitative Researcher (Rates)
Humit2021
Founding Backend Engineer
RIKEN, Japan2019-20
Research Assistant (Franco Nori Group)
tensoralpha GmbH2020
Quantitative Developer Intern (Thomas Schmelzer)
Microsoft Research2019
Research Intern
IISER Pune2018-19
Undergraduate Thesis (Prof. Sreejith GJ)
WorldQuant2018-19
Research Consultant
ISI Calcutta2018
Cryptology Research Intern (ACM Summer School)
Zero Labs2017
R&D Intern (NLP)
Research Projects
Exceptional Points in Non-Hermitian Quantum Systems
Studied exceptional points and exotic properties of large non-Hermitian systems using a novel algorithm I developed
Noisy Mean Field Annealing (NMFA) for Combinatorial Optimization
Optimized NMFA for NP-hard problems using ideas from quantum thermodynamics
Quantum Overlapping Tomography
Developed quantum overlapping tomography module using perfect hash families based on Frank Wilczek's paper; studied quality of entanglement in IBM's machines
PyFroid
Plugin to Pandas; achieved 100x speedup on certain operations through lazy evaluation
Ontology-Aware Medical Concept Extraction
BiLSTM-CRF and Transformers with UMLS embeddings for EHR named entity recognition
Speech Event Segmentation
Advanced metrics for speech segmentation via quantile analysis of intersyllable durations; stochastic point processes and neural equivalents for syllable event modeling
Neural Network Potentials for Molecular Dynamics
ResNet-based NNPs for copper thermal expansion, validated with LAMMPS
Quantum & Classical Key Exchange Protocols
Selected among top 50 undergrads nationwide for cryptology research program
Computational Geometry Library
C++ implementations of convex hull algorithms, polygon triangulation, point visibility
MPI Parallel Code Optimization
PETSc-based optimization for aerofoil sensitivity analysis on GPU-CPU hybrid systems
Data Envelopment Analysis for Student Performance
DEA-based efficiency analysis of student input factors
Blogs & Talks
Cisco Outshift Blog (2024)
An Initial Exploration of Using Persistent Homology for Noise-Resilient Tactile Object Recognition
Oral Presentation, NeurIPS 2024 Workshop on Touch Processing (Dec 2024)
Reviewing
Advisors
I've been fortunate to work with some truly amazing people.
Andrea Liu (UPenn) ·
Mark Liberman (UPenn) ·
Franco Nori (RIKEN) ·
Neill Lambert (RIKEN) ·
Karthik Ramachandra (MSR) ·
Thomas Schmelzer (tensoralpha) ·
Hidenori Tanaka (Harvard/NTT) ·
Fatih Dinc (Stanford) ·
Menachem Stern (UPenn) ·
Sreejith GJ (IISER Pune) ·
Jayanth Srinivasa (Cisco Research)
Misc
I wrote the algorithm for WeirdSpotify with Humit - it went viral. Favorite read: HPMOR.
Will add more stuff here.