Gurarmaan S. Panjeta

Quantitative Researcher @ Trexquant | M.Tech & B.Tech in CS + Physics Minor @ IITD '25

resume.pdf github.com/panjete linkedin email

// Core abstractions I've wrestled with so far:

// Quant · Portfolio Optimization

$$\max_{w} \left( w^T \mu - \frac{\gamma}{2} w^T \Sigma w \right)$$

// Machine Learning · Policy Gradient RL

$$\nabla_\theta J(\theta) = \mathbb{E}_{\pi_\theta} \left[ \nabla_\theta \log \pi_\theta(a|s) Q^\pi(s,a) \right]$$

// Physics · Quantum Dynamics

$$i\hbar \frac{\partial}{\partial t} \psi(\mathbf{r}, t) = \hat{H} \psi(\mathbf{r}, t)$$

// Information Theory · VAE ELBO

$$\mathcal{L}_{\text{ELBO}} = \mathbb{E}_{q_\phi}\left[\log p_\theta(x|z)\right] - D_{\text{KL}}(q_\phi \parallel p)$$

"Citius, Altius, Fortius"

Publications & Reports

[ICML '25] · PRAL Workshop

Sketch-Plan-Generalize: Enabling Continual Learning of Inductive Programs for Spatial Concepts

Enabling continual learning of inductive programs for spatial concepts from human demonstrations.

[DARIAH-DH '25] · Conference Paper

Tracing Temporal Language Dynamics in Media via LLMs

Identifying correlations of historical events with cultural trends by dissecting popular media via Large Language Models.

[Master's Thesis] · Spatial Reasoning & Scene Maps

Neuro-Symbolic Demonstration-Aligned Continual Learning & Generalized Inference of Inductive Spatial Concepts

Advised by Prof. Rohan Paul & Prof. Parag Singla. Developed language-aided concept-conditioned voxelised spatial maps for object placement and scene reasoning, disentangling spatial reasoning from agent controls via scene-conditioned probability distributions.

[PROJECT REPORT] · Astrophysics & Spectroscopy

Investigating Spatially Resolved CGM Tracers in Nearby Galaxies Using PHANGS-MUSE

Advised by Prof. Suprit Singh. Derived NaI & CaII spatial flux distributions in PHANGS-MUSE Galaxies via Voronoi Binning & Gaussian fitting.

## TL;DR; "(logging key checkpoints)"

Presented @ DARIAH-HR 2025, Croatia

Presented paper "Tracing Temporal Language Dynamics in Media via LLMs" in Osijek, Croatia, dissecting popular media to identify and model correlations between historical events and cultural trends over time.

Graduated @ IIT Delhi

Conferred degrees from IIT Delhi:

M.Tech in Computer Science (9.804/10 GPA)
B.Tech in Computer Science (9.008/10 GPA)
Minor Degree in Theoretical and Computational Techniques in Physics

Presented @ PRAL Workshop @ ICML 2025, Vancouver

Attended the conference in person to present our accepted work on Neuro-Symbolic AI . Inspirational to hear from and meet both industry and academic giants in domains like RL, time-series, and flow-based models.

Outstanding Teaching Assistantship Award (2x Winner)

Conferred Outstanding TA Award twice across consecutive semesters by the CSE Department for contributions to COL778: Principles of Autonomous Systems (Head TA) and COL333: Principles of Artificial Intelligence (TA). View Department Announcements .

Quantitative Research Intern: Trexquant

Involved in Alpha Research and using Machine Learning techniques to tune financial signals, Monte Carlo feature selection, short-term momentum Alphas, and signal optimization.

Teaching Assistant: COL100 (Intro to CS)

Served as a Teaching Assistant managing labs and tutorial sessions, resolving assignment doubts, and evaluating answer scripts for freshmen.

MITACS Research Intern: Machine Learning & Optimization

Spent 10 weeks at the University of Western Ontario, Canada, optimizing Time To Accident (TTA) prediction using dashcams. Meticulously analyzed and tuned loss functions for reliable training.

Scholastic Achievements

2025 Outstanding Teaching Assistantship Award — Conferred for contributions as Head TA to COL778: Principles of Autonomous Systems.
2025 Tower Research Capital / Limestone Data Challenge — Finished 2nd place nationwide, awarded 150k INR.
2024 Outstanding Teaching Assistantship Award — Conferred by CS&E Dept. for contributions to COL333: Principles of Artificial Intelligence.
2023 MITACS GRI Scholar — Awarded Globalinks Research Internship scholarship at UWO, Canada.
2022 Ranked 8th, Foreign Exchange — Selected based on merit & interpersonal skills to represent IIT Delhi abroad.
2021 IITD Academic Merit Scholarship — Awarded to the top 7% cohort for Ist & IInd semesters.
2020 All India Rank 115, JEE Advanced — Amongst 250,000 aspirants in the national entrance exam.
2020 All India Rank 286, JEE Mains — Out of 1.1 million aspirants nationwide.
2020 Indian Physics Olympiad (InPhO) — Highest National Honor in Physics (Top 35 nationwide for school students).
2020 All India Rank 259, KVPY Fellowship — Aced the Research Fellowship Exam & Interview by IISc Bangalore.
2019 National Standard Examinations (NSE) — Top 1% nationwide in Physics, Chemistry & Astronomy; selected for InPhO / InChO / InAO.
2018 NTSE Scholar (National Talent Search Examination) — Conferred by GoI to top 800 out of 1 million examinees.

Projects

I dabble with fields like ML, RL, Autonomous Systems, Representation Learning, LLMs, Deep Learning, Linear Algebra, Computer Networks, Circum-Galactic Matter, the Humanities (and much more). PFA a non-exhaustive subset:

Kalman Estimator

State Estimation using Kalman Filters, EKF, and solving the Data Association problem. Simulations, plots, and analysis across noisy environments.

$$\hat{x}_{k|k} = \hat{x}_{k|k-1} + K_k \left(z_k - H_k \hat{x}_{k|k-1}\right)$$

Inverted-Index Search Engine

Retrieval system searching 250K documents (1MB each) in 1.12s. Incorporates Variable Byte Compression and Byte Pair Encoding.

$$\text{Latency} \le 1.12\text{s} \quad \text{for } 250\text{K docs}$$

AI & RL Ventures

Search algorithms, state estimation, planning tasks, and RL via automated drivers, English Error-Corrector, MuJoCo Gym agents, and Actor-Critics.

$$\nabla_\theta J(\theta) = \mathbb{E} \left[ \nabla_\theta \log \pi_\theta(a|s) Q^{\pi}(s,a) \right]$$

Re-Rank Documents

Evaluating pseudo-relevance feedback, query expansion, and re-ranking models aimed at improving precision/recall on TREC COVID dataset.

$$\text{MAP@K} \quad \& \quad \text{NDCG@K}$$

OS: Linux Schedulers & Modules

Augmented xv6 kernel Round Robin with EDF and RMS schedulers, plus virtual Reader & Writer devices using Linux kernel modules.

$$U = \sum_{i=1}^n \frac{C_i}{T_i} \le n \left(2^{1/n} - 1\right)$$

P2P Network & TCP Simulations

TCP NewRenoPlus Congestion Control Protocol implementation and simulation of a variable-cached peer-to-peer network topology.

$$w \leftarrow w + \frac{1}{w} \quad (\text{Additive Increase})$$

Contact

Have a question, research inquiry, or just want to get in touch? Leave a message below or email me directly at suparmaansingh@gmail.com.