Hello! I'm Nimish Jain, a Computer Science student at Vellore Institute of Technology working at the intersection of AI research and engineering, driven by a passion for building intelligent systems that solve tangible, real-world problems.
My core philosophy is to bridge the gap between theoretical models and practical, deployable solutions. I have hands-on experience across the entire machine learning lifecycle—from data preprocessing and feature engineering to model architecture design, training, and performance optimization. I thrive on tackling problems that require not just technical skill but also a deep understanding of the problem domain.
I'm committed to pushing the boundaries of what's possible, one robust and thoughtful solution at a time.
Future Goals
My ambition is to be at the forefront of AI innovation—to create and to invent. I am driven to deepen my expertise in machine learning, immerse myself in cutting-edge research, and ultimately contribute to developing novel systems and technologies that push the boundaries of what's possible.
Get in Touch
jainnimish245@gmail.com
Research Experience
Research Intern
Indian Space Research Organization, URSC | ISITE, Bangalore
May 2025 - July 2025
Under the guidance of esteemed scientists at ISRO, my research focused on a critical bottleneck in interplanetary navigation: the high-variance, unpredictable nature of atmospheric density during spacecraft aerobraking. We identified that conventional static models introduce significant risk and are insufficient for ensuring mission safety. My primary contribution was the design and implementation of a lightweight, physics-guided machine learning model for real-time density estimation. By integrating an adaptive Kalman Filter, the system continuously learns from noisy, in-situ data, significantly enhancing prediction accuracy and robustness. This work represents a foundational step toward developing fully autonomous navigation systems capable of operating in uncertain extraterrestrial environments.
Research Intern
IIT Madras Research Park | Remote
December 2024 - May 2025
In collaboration with researchers at IIT Madras, I tackled the complex challenge of regional water quality management. My research involved the analysis of a massive dataset, comprising over 1 million data points from the Haryana government, which included critical parameters like pH, nitrogen, and fluoride levels across more than 20 urban centers. I engineered a sophisticated predictive model to forecast water purity in real-time and determine necessary treatment levels. Through a rigorous process of feature engineering, hyperparameter optimization, and iterative evaluation, the resulting system provides a scalable, data-driven tool for environmental governance with direct applications in public health.
Research Papers
Bridging Neural and Symbolic Reasoning: Advances in Hybrid AI
Pre-print: April 2025
My curiosity is intensely focused on the frontiers of AI, especially in making large language models more interpretable and trustworthy. This conviction led me to the field of Neuro-Symbolic AI. While deep learning excels at pattern recognition, it often operates as a 'black box.' In contrast, symbolic AI provides clear, logical reasoning but lacks adaptability. In this paper, my co-authors and I investigated how these two paradigms could be effectively bridged.
We explored the integration of symbolic AI's reasoning capabilities with the pattern-recognition strengths of neural networks to build systems that can both learn from diverse data and reason about abstract concepts. Our goal is to contribute to an AI that is not only powerful but also transparent, reliable, and less susceptible to bias. This research reflects my commitment to advancing the field toward more robust and explainable artificial intelligence.
Read Full Paper →Blog Posts
Strawberry Model vs Its Predecessors: What’s New in AI?
Here, I use the 'strawberry' problem not just as a novelty, but as a case study to dissect the architectural limitations of LLMs. My analysis moves from identifying the tokenization flaw in earlier models to a deeper investigation of the 'Chain of Thought' mechanism in OpenAI's `o1`. I explore how reinforcement learning likely underpins this shift, demonstrating a move from simple pattern matching to a more structured, sequential reasoning process—a critical evolution for building more reliable AI.
Read on Medium →Beyond the Title: Understanding Apple’s ‘Illusion of Thinking’ Paper
A provocative title demands rigorous scrutiny. In this post, I deconstruct Apple's paper on Large Reasoning Models (LRMs) by questioning its core methodology. My analysis focuses on the use of classic puzzles for evaluation, which risks data contamination and may compromise the results. I probe the paper's conclusions, distinguishing between genuine model limitations and potential artifacts of their experimental design. It's my attempt to look past the hype and engage with the fundamental questions the research raises about how we truly measure progress in AI reasoning.
Read on Medium →Featured Project (Patent Pending)
SafeSite AI: A Multi-Stage System for PPE Compliance Verification
My work on SafeSite AI began with a critical analysis of existing automated safety systems, which I found were often limited to simple object detection. This leads to a critical flaw: a high rate of false positives where the mere presence of an item, like a helmet on the ground, is incorrectly flagged as compliant. To address this, I engineered a more robust paradigm for safety monitoring.
I designed a multi-stage validation pipeline that moves beyond presence detection to verify correct usage. The system integrates a YOLOv8 model for multi-class detection, instance segmentation to resolve ambiguities in crowded environments, and a novel rule-based engine for positional verification. This approach doesn’t just identify PPE; it contextualizes it, translating raw visual data into actionable safety intelligence. This project, currently undergoing the patent process, represents my commitment to building systems that solve real-world challenges through deep, domain-aware AI.
Letters of Recommendation
This section is currently under construction.
Please check back later for more information.
This space is dedicated to my broader engagement with the tech community—from workshops I've led to viewpoints I've shared on various platforms.
Interactive AI Education: Deconstructing Neural Networks
Beyond my formal research, I'm passionate about making complex AI concepts accessible and intuitive. As the Events Head for my university club, I conceived and led a flagship session for our tech festival, Gravitas, merging mental health, neuroscience, and AI. My goal was to move past conventional lectures and demonstrate my core understanding of neural networks through creative pedagogy.
I designed an interactive, 3D "human neural network" game where participants became the neurons, ropes symbolized connections, and cards represented weights. This experiential approach was my solution for translating abstract theories—like weights, biases, and propagation—into tangible, memorable lessons. This initiative showcases my leadership in developing engaging educational formats and my ability to distill and communicate foundational AI principles creatively.