Physics undergraduate at NIT Agartala, working toward research in machine learning, continual learning, and neuromorphic-inspired computation.
I like building systems from first principles and testing them honestly: what works, what fails, and which claims the evidence actually supports. I am currently looking for research opportunities where I can strengthen my ML foundations while contributing through Python, experimentation, documentation, and software engineering.
MNEMA — continual-learning research prototype
A NumPy implementation combining sparse representations, fast associative memory, a shallow spiking-style learner, and consolidation.
- Evaluated on Split-MNIST Class-IL against five continual-learning baselines
- Completed 60 runs: six methods across ten random seeds
- MNEMA: 77.12 ± 0.87% final accuracy and 6.53 ± 1.41 pp forgetting
- Includes reproducible artifacts, explicit memory/energy boundaries, and a technical report
- Technical report · Benchmark results
A chess engine with complete legal move generation, iterative-deepening alpha-beta search, transposition tables, UCI support, SAN/PGN notation, benchmarks, a Tkinter GUI, CI, and 83 tests.
An end-to-end ML application using sequential ELO ratings, leak-free rolling features, calibrated XGBoost probabilities, and Monte Carlo tournament simulation. The model reaches approximately 57% accuracy on a chronological three-class holdout.
- Continual learning and catastrophic forgetting
- Neuromorphic and memory-centric learning systems
- Representation learning and neural architectures
- Scientific computing and physics-informed approaches to AI
- Reproducible experimentation and model evaluation
- Languages: Python, C, JavaScript, TypeScript, SQL
- ML and data: NumPy, pandas, scikit-learn, PyTorch, Matplotlib
- Software: Git, GitHub Actions, Linux, Docker, REST APIs
- Web: React, Next.js, Node.js, Express, Tailwind CSS
- Data systems: PostgreSQL, MongoDB, Redis, Prisma
Integrated BS-MS in Physics
National Institute of Technology Agartala · 2025–present
- Email: ujandey007@gmail.com
- Portfolio: ujandey.vercel.app
- GitHub: @ujandey