A local research workspace for traceable tensile analysis: from machine exports to versioned results, specimen comparison, and reproducible reports.
Hello, I’m
Ankit Kumar.
AI engineer & software developer
I work on language models, reinforcement learning, and backend systems. I also build tools for my Linux desktop and enjoy understanding how things work from the ground up.
Experience acrossAI research at IIT GuwahatiReinforcement learning at ISROProduction backend development
Selected work
Projects I’ve worked on
A selection of work across machine learning, backend engineering, and scientific computing.
Building GPT-style transformers from their individual components, then adapting them for instructions and spam classification.
Q-learning and deep reinforcement learning in FrozenLake, CartPole, and a custom Tank Dodge environment.
A Django web application bringing multiple machine learning architectures into one model-evaluation interface.
A keyboard-driven project launcher with fuzzy search, Git status, editor actions, and automatic task discovery.
A ticket-management backend with role permissions, searchable endpoints, and asynchronous escalation tasks.
Experience
Where I’ve worked
Indian Institute of Technology, GuwahatiAI/ML ResearchJun 2026 - Sep 2026
- Conducted research on next-generation AI, Machine Learning, and Large Language Model (LLM) architectures with a focus on improving efficiency, reasoning, and scalability.
- Explored state-of-the-art Transformer architectures, multimodal models, Mixture of Experts (MoE), diffusion models, and emerging foundation model techniques.
- Designed and implemented deep learning models using PyTorch and Python for experimentation and evaluation.
Freelance ConsultingFreelance AI/ML Engineer | ConsultantJuly 2024 – Present
- Designed and deployed 4+ production-grade Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) systems for domain-specific applications, achieving a 20% improvement in search/retrieval performance.
- Built and fine-tuned transformer-based Large Language Models (LLMs), including training a 124M parameter model from scratch.
- Designed and implemented end-to-end conversational AI solutions that scaled to handle 5,000+ daily user interactions.
- Reduced harmful output instances by 95% through the implementation of AI safety guardrails and advanced prompt engineering techniques.
ISRO - Indian Space Research OrganisationProject InternMay 2024 – July 2024
- Optimized autonomous Lunar landing simulations using Reinforcement Learning agents, reducing trajectory error margin by 15% over baseline heuristics.
- Engineered Range Operations data pipelines to process 10,000+ data points per second, enabling real-time fault forecasting with 90% accuracy.
- Designed a low-level hardware processing layer in Python/C to increase telemetry throughput by 2X and condition GBs of multi-sensor telemetry for RL training loops.
- Collaborated closely with a team of distinguished Scientists to prototype mission-critical software frameworks.
EcowiserBackend Developer InternNov 2022 - Apr 2023
- Designed and scaled production backend services for a core Chrome Extension and Web ecosystem supporting 10,000+ active users utilizing Django Rest Framework.
- Accelerated internal search query speeds by 30% through the integration of custom NLP tokenization into PostgreSQL Full-Text Search.
- Built an asynchronous, automated customer engagement pipeline utilizing Redis and Celery, processing over 5,000 scheduled feedback dispatches daily.
Velozity Global SolutionsMachine Learning InternJune 2022 - Oct 2022
- Developed and cross-validated predictive deep learning frameworks using PyTorch and TensorFlow across 15GB+ unstructured datasets.
- Designed and benchmarked LSTM-based architectures, improving accuracy by 14% in identifying long-term temporal dependencies within sequential data logs.
About
A little about me
I’m a software engineer interested in machine learning, especially language models, computer vision, and reinforcement learning.
Alongside research and application development, I use Arch Linux(btw) and build small tools to improve my everyday workflow. My GitHub includes finished applications, experiments, and notes from learning new things.
Outside workB.Tech in Computer Science and Data Science
Lovely Professional University · 2020–2024
Machine learning
PyTorch · TensorFlow · OpenCV
Language models · Retrieval · Computer vision · Reinforcement learning
Software engineering
Python · C++ · SQL
FastAPI · Django · PostgreSQL · Redis · Celery
Shipping & experimentation
Docker · Linux · Git · MLOps
Streamlit · SciPy · Pandas
Writing
Articles & notes
Attention Is All You Need, Explained from the Ground Up
A visual guide to the original Transformer: intuition first, then attention mathematics, tensor shapes, masking, and small PyTorch examples.
Tensile Strength Analysis
A comprehensive guide to tensile testing analysis, covering stress-strain relationships, Young's Modulus, Yield Strength, and how machine learning techniques improve material property calculations.
Understanding Large Language Models: Mathematics and Architecture
A deep dive into the mathematics and architecture behind Large Language Models, from transformers to attention mechanisms.
Contact
Let’s talk
If you’re hiring for machine learning or software engineering, I’d be happy to hear about your team and the role.
nasaankit01@gmail.com