HomeAboutSkillsProjectsExperienceContact

Hello, I'm

Durga Sai

B.Tech Computer Science Student at IIT Kharagpur

Junior at IIT Kharagpur, Class of 2028, building machine learning systems through neural architecture research, efficient LLM inference, and multi-agent simulations.

View ProjectsDownload Resume
Scroll

About Me

I am a Junior at IIT Kharagpur pursuing a B.Tech. (Hons.) in Computer Science and Engineering, with a micro-specialization in Artificial Intelligence and Applications, Class of 2028. I build machine learning systems from the underlying mathematics through to usable software: matrix-valued neural networks in JAX, LLM inference on Intel NPUs, and multi-agent simulations. My work includes MATNETS, NERVI, and an IPL auction simulation under Prof. Pawan Goyal. Alongside research and open-source development, I have worked on automation as a software development intern at Asandevnest and explored retrieval-augmented reasoning as a KDSH Hackathon finalist. I enjoy connecting neural architecture design, efficient computation, and practical tools.

Junior at IIT Kharagpur · B.Tech Computer Science · Class of 2028

🎓
Education
B.Tech CS, IIT Kharagpur (2024–2028)
📍
Location
Kharagpur, India
🧠
Interests
AI/ML, LLM Inference, Neural Architectures
🔬
Research Focus
Neural Architectures & Algorithms

Skills & Technologies

⟨/⟩

Languages

PythonC++GoTypeScriptJavaScript
◎

AI & Neural Networks

Deep LearningNeural Network ArchitectureMulti-Agent SystemsModel Fine-TuningComputer Vision
≡

LLM Inference & Optimization

OpenVINOQLoRAINT4 QuantizationIntel NPU Inference
⚙

Systems & Infrastructure

Linux & GitHigh-Performance SystemsCloudflare WorkersAutomation Frameworks

Featured Projects

A curated showcase of research and software — matrix-valued neural networks, efficient LLM inference, multi-agent simulations, and practical tools.

01

MATNETS

Matrix-Neuron Neural Networks library in JAX redefining the neuron as an n x n matrix. Features determinant-gated activation families, matrix-native pooling, and Frobenius-scored Attention. Published on PyPI.

JAXFlaxPythonNeural Network Design
View Source
02

NERVI

Intel NPU LLM inference ecosystem achieving 13.1 tok/s on DeepSeek-R1-1.5B with no GPU. Features QLoRA fine-tuning, INT4 quantization, OpenVINO conversion, and an OpenAI-compatible PyPI server.

GoPythonOpenVINOLLM Serving
View Source
03

Multi-Agent IPL Auction Simulation

Research project under Prof. Pawan Goyal (IIT Kharagpur). Built a 5-node LangGraph pipeline with 10 DeepSeek-V3.1 franchise agents simulating IPL mini-auctions over 370+ players.

PythonLangGraphLangChainMulti-Agent AI
View Source
04

ENIGMA

Custom C++17 password hashing & encryption engine with pybind11 Python bindings, GIL release, and parallel encryption at 50 MB/s on 4 threads. Published wheels to PyPI.

C++17PythonPyPICryptography
View Source
05

Electric Charge Distribution

Constrained gradient descent over 10,000 Fibonacci-lattice points with mean-centred gradient projection achieving ~97% similarity against exact Legendre series solutions.

PythonNumPyPandasPlotly 3D
GitHub Profile
View All on GitHub

Experience

Asandevnest

Software Development Intern

Jun 2025 – Sep 2025

PythonSeleniumPyWhatKitN8NApifyAutomation
Remote

Get In Touch

Let's work together

I'm always open to discussing machine learning research, neural architectures, efficient LLM inference, or software engineering collaborations.

Kharagpur, West Bengal, India