/now
Currently in the lab.
- ResearchingReasoning-model failure modes (7B to 70B)
- StackPyTorch · Slurm · 4× GH200
- Side benchGARCH(1,1) on SPY, walk-forward
- Based inDayton, Ohio
- Emailabhijeetguptaphd@gmail.com
PhD · University of Dayton
Abhijeet Gupta. — ML Engineer & Quant Researcher
signal · t = now
α = 0.07
Currently exploring
Manufacturing AutomationAGILLM Interpretability
▲LLM-EVAL6 models · 7B to 70B
▲ADV-RL94% acc · KSE'24
▲PAIRSAAPL-MSFT · Sharpe 1.8
▲GARCH(1,1)SPY · σ̂ live
▲OPTIONSMC · 100k paths
▲CVCISS'25 ×2
▲PUBS4 peer-reviewed
▲STACKPyTorch · Slurm · 4×GH200
▲LLM-EVAL6 models · 7B to 70B
▲ADV-RL94% acc · KSE'24
▲PAIRSAAPL-MSFT · Sharpe 1.8
▲GARCH(1,1)SPY · σ̂ live
▲OPTIONSMC · 100k paths
▲CVCISS'25 ×2
▲PUBS4 peer-reviewed
▲STACKPyTorch · Slurm · 4×GH200
Two disciplines, one practice
Research and markets.
ML / AI Research
LLM evaluation, reinforcement learning robustness, and computer vision pipelines on multi-GPU HPC infrastructure.
Apr 2026
Reasoning Model Failure Analysis, LLM Interpretability
KSE 2024, Published
Adversarial Robustness via Entropy Based Feature Selection in RL
Quantitative Research
Options pricing, GARCH volatility modeling, and market-neutral statistical arbitrage.
More →Currently reading and considering
Manufacturing Automation
Integrating reproducible machine learning and statistical process control into legacy production lines, where the operational bottleneck is rarely the model itself.
AGI
Examining what 'general' substantively means once it is operationalized for measurement. Currently reading critiques of scaling laws and recent work on evaluation design.
LLM Interpretability
Investigating reasoning-model failure modes, family-level differences between Llama and Qwen, and the conditions under which forced re-entry helps or harms performance.
Most recent publication: “Enhancing Sustainability and Construction Safety Research in the Era of Artificial Intelligence,” ASME Journal of Engineering for Sustainable Buildings and Cities 2026. View all →
Explore the work
Browse by skill.
Every project I have shipped, tagged by the technologies it employs. Select a skill to see where it has been applied.
Project explorer · interactive
Click a skill, see what I've built with it.
Showing
8 / 8
- ML / AIApr 2026
Reasoning Model Failure Analysis, LLM Interpretability
PyTorchHuggingFaceSlurmMulti-GPU - ML / AIKSE 2024, Published
Adversarial Robustness via Entropy Based Feature Selection in RL
OpenAI GymPyTorchRLPublication - ML / AICISS 2025, Published
Mouse Brain Cell Segmentation in Fluorescence Microscopy
OpenCVPyTorchCNNsPublication - ML / AICISS 2025, Published
Virtual Yoga Instructor with Real Time Feedback
OpenCVMediaPipePyTorchPublication - QuantJan 2026
Options Pricing Engine and Greeks Computation
PythonNumPySciPyMonte Carlo - QuantDec 2025
Statistical Pairs Trading Backtest
PythonstatsmodelsyfinanceBacktesting - QuantMay 2025
GARCH Volatility Modeling and Stochastic Time Series
PythonstatsmodelsGARCHTime Series - QuantDec 2023
LSTM Based Financial Time Series Forecasting
PyTorchLSTMTime Seriesyfinance
20 skills · 8 projects · click to filter
Writing
Notes and essays.
Occasional writing on research, tooling, and learning in public.
All writing →Jun 12, 2026
How to correctly report LLM-as-a-Judge evaluations
A practical guide to running, calibrating, and reporting LLM-as-a-Judge results — covering judge selection, position bias, pairwise vs scoring setups, and the statistics that actually belong in the paper.
Jun 10, 2026
10 must-read machine learning research papers for ML engineers
An annotated bibliography of foundational and recent work in LLM evaluation and reinforcement learning, with notes on why each paper matters in practice.
Apr 20, 2026
Notes on evaluating reasoning models across families
Observations from disentangling reasoning length effects from forced re-entry across Llama and Qwen distilled models.
Nov 2, 2025
Reproducibility on a shared Slurm cluster
Small operational habits that yield significant returns when several collaborators share the same GPU resources.
Get in touch