Hello, I'm Eswar Gupta
Undergraduate in Electrical Engineering at IIT Madras.
Software Engineer · AI/ML · Competitive Programmer
Achievements
Ranks: Secured AIR 737 in JEE Mains (out of 1.3 million) and AIR 2167 in JEE Advanced (Top 0.5%).
Programming: Codeforces Specialist (1451 rating) [Handle: Eswar_Gupta], ranked 644 in Div2 (45k+ participants).
Other Competitions: Secured 20th Position in the 13th National Abacus & Mental Arithmetic Competition (10,000+ participants).
Skills
Languages & Frameworks
C, C++, Python, SQL, MATLAB, FastAPI, Spring Boot, Pydantic, MLFlow, PyTorch, OpenAPI, Verilog, AVR, ARM, Arduino, NumPy, pandas, Matplotlib, cython, NLTK, Flask
Tools & Platforms
PostgreSQL, Postman, gCloud, Git, Teleport, Docker, Kubernetes, Jupyter Lab, GitHub, Notion, LaTeX, Vivado, Render
Relevant Coursework
- Data Structures & Algorithms
- Machine Learning
- Database Management Systems
- Deep Learning
- Computer Networks
- Reinforcement Learning
- Computer Organization
- Andrew Ng ML Specialization
- C & Python Programming
- Probability
- Applied Programming (Python)
- Functions of Several Variables
- Digital Systems
- Series & Matrices
- Microprocessor Theory & Lab
- Signals & Systems
- Control Systems
- Digital Signal Processing
Professional Experience
Built a scalable Agent Evaluation Platform engine to support multiple autonomous AI agents across different teams of JPMC.
- Developed an offline and online evaluation engine from scratch to monitor AI agents pre- and post-deployment, integrating 50+ distinct LLM and deterministic metrics utilizing a Hexagonal Spring Boot orchestrator and Python microservices.
- Reduced trace evaluation latency from 130s in MVP to under 30s by iteratively optimizing the architecture, and engineered a dynamic configuration API with a robust OpenAPI contract allowing developers to seamlessly configure metrics.
- Built custom unsupervised evaluators (Hallucination, Tool Call Efficiency) and leveraged CockroachDB and MLflow servers for trace management, deployed this product using Docker, Kubernetes, and Helm.
Implemented an AI audio analyzer (Python engine) and integrated it with a dashboard to visualize audio features.
- Implemented OpenAI APIs for baseline inference, subsequently training and deploying a lower-quality, cost-effective WavLM model on EC2 to drastically reduce compute expenses while extracting 56 distinct linguistic features.
- Benchmarked WavLM against ExHUBERT through a SWOT analysis, finalizing WavLM as the optimal local model.
- Proposed and developed a training data pipeline to foolproof the Artsens neurovascular instrument by converting ultrasound signals into B-mode images and using YOLOv8 for vessel detection.
- Engineered an overlap control feature to optimize the number of images generated per patient scan, and implemented smart deadband filters to clear disturbances from gel applied to patient's neck, ensuring high-quality B-mode images.
- Leveraged YOLOv8 and wall-motion to detect blood vessels, processing 4000 patients data and generating 1 million images.
Projects
Delta Hedging Simulator: Black-Scholes vs DL
- Engineered a minute-level discrete hedging simulator, generating 10,000+ synthetic GBM market scenarios to train an RNN for dynamic European options hedging.
- Benchmarked the neural-network-based Deep Hedger against classical Black-Scholes, evaluating risk metrics including VaR, CVaR, and P&L standard deviation.
- RNN improved average P&L by 7.7% but suffered 2.2x worse CVaR, suggesting neural networks require adversarial architectures for real-world market crashes.
Low-latency C++ Memory Allocator
- Engineered a Low-latency C++ memory allocator utilizing an O(1) free-list and a type-safe ObjectPool, eliminating dynamic heap allocation to process 1M trade orders in ~39ms (6.5x latency reduction vs standard new/delete).
- Designed an STL-compatible PoolAllocator to back node-based order book structures, tripling node insertion throughput (3.2x speedup) by efficiently packing nodes into 64-byte cache lines.
Giteswar.com
- Engineered a GitHub utility for repository/subdirectory analysis and targeted downloading. Replacing github.com with giteswar.com opens an inspection dashboard, bypassing manual terminal-based Git sparse-checkouts.
- Leveraged this repository analysis to power an LLM Context Generator that aggregates directory structures, tech stacks, and key files across subdirectories to feed project context into LLMs.
- Architected the application using Python and FastAPI, containerized with Docker, and deployed on Google Cloud Run.
Cython Performance Analyzer
- Engineered Cython extensions for Monte Carlo pricing, completely eliminating Python interpreter overhead to achieve 47x speedups through static C type definitions.
Electronics Club SONIC
- Real‑time noise filtering for crowds, adapting filter coefficients to ambient noise.
- LMS‑based dual‑mic algorithm in MATLAB → deployed on STM32 in embedded C.
- Correlated mic‑2 noise‑only signal with mic‑1 to isolate speech.
- Wrote embedded‑C WAV encode/decode for end‑to‑end audio processing.
Heart Disease Tracker
Analysed heart disease dataset to predict patient risk; goal was accurate binary classification using health parameters. Trained Neural Network, KNN, Random Forest, xgboost algorithms and Random Forest performed best with 89% accuracy.
Applied Programming Lab
- Simulated circuit‑solver: node voltages, branch currents & power from .ckt files.
- Python scripts: keyboard‑usage analysis, heatmap generation, finger‑travel‑distance calc.
- Simulated annealing to optimize keyboard layouts; animated results via Matplotlib.
- DAS acoustic imaging: obstacle detection & effect of mic spacing/sampling rate.
Machine Learning Fundamentals
- Implemented core ML algorithms from scratch: Linear/Logistic Regression, Multiclass Classification, and Neural Networks
- Applied advanced practices: model evaluation, bias/variance analysis, regularization, and learning curves
- Built unsupervised learning algorithms including K-means clustering and anomaly detection systems
- Developed NLP applications using NLTK, implementing text preprocessing, BOW, TF-IDF, and Word2Vec embeddings
Movie Recommender Systems
Engineered comprehensive movie recommender systems using datasets like Movie-Lens, implementing collaborative filtering from scratch and leveraging neural networks for content-based filtering to personalize user recommendations.
Positions of Responsibility & Social Impact
Department Legislator & Class Rep (Mar '26 – Mar '27)
- Elected with an 82% majority to represent the EE department, coordinating a department trip with a ₹7.5 lakh budget.
- Selected for a 5-day trip in collaboration with the Government of India to visit Parliament and engage with multiple ministers, having also served as Class Representative for 3 years.
Electronics Club Coordinator
- Organized a Summer School with 10,000+ students across 14 verticals, coordinating and teaching technical sessions.
- Guided 200+ students in building hardware projects using Arduino and ESP32 across multiple offline workshops.
Avanti Fellows Mentor
- Mentored underprivileged students through India's largest STEM education and entrance exam mentoring NGO.
Teaching Assistant
- Serving as problem setter and mentor for Programming in C lab.
About Me
I am an Electrical Engineering student at IIT Madras, with a background from Bhashyam College. Originally from Andhra Pradesh, I am a proactive individual who readily learns from others and adapts to new technologies and environments.
My interests span mathematics, technology, signal processing, AI, and communications. As an extrovert, I enjoy connecting with new people, and I have a keen interest in news and social welfare activities.
Get in Touch
I'm always open to new opportunities, collaborations, or just a friendly chat. Feel free to reach out via any of the methods below.