Shuvam Banerji Seal — Computational Chemist × AI Researcher
I work where molecular simulation meets the web. I've published retrieval research at TREC, modelled catalysts with DFT, and built browser-native tools that scientists actually use — a GPU-accelerated structure viewer, a browser DAW, a 448-track player. Not just functional — crafted.
I'm Shuvam — a BS-MS student at IISER Kolkata pursuing Chemistry with a Computer Science minor. My research focuses on computational chemistry, information retrieval systems, and AI-driven solutions for scientific challenges. I've developed advanced RAG architectures for verifiable answer synthesis, engineered retrieval models for TREC competitions, and conducted DFT computational modeling of catalysts.
My work spans molecular dynamics simulations to full-stack application development using Python, Java, C, and modern web frameworks. I'm passionate about interdisciplinary research that bridges computational methods with scientific discovery.
// Computational
// Programming
// AI / ML
// Tools
GPU-accelerated 3D molecular structure viewer and LAMMPS workbench: 186-command visual script builder, CMake compiler helper, dump-trajectory playback, measurement tools — four structure formats, entirely in the browser.
Comprehensive RAG pipeline for IISER-K intranet with automated data acquisition, OCR processing, and hybrid search combining dense vector retrieval with BM25 for verifiable answer synthesis.
Search that works when you only remember how the word feels. Multi-layer BM25 filtering, dynamic search-domain contraction, and local-LLM query expansion — published in the TREC 2024 proceedings.
DFT methods (B3LYP) for MO energy calculations and electronic structure analysis of [VO(SALIEP)(DTP)] as water-reducing catalyst using Gaussian 16 and transition state theory.
Complete GUI application in C using GTK4 and Glade, with SQLite3 database management and optimized BM25-based retrieval for NLP chatbot functionality on desktop Linux.
Computational analysis of protein-ligand interactions and conformational dynamics using molecular dynamics simulations, with visualization pipelines built on PyMOL and custom Python tooling.
Classical MD, end to end.
GROMACS and LAMMPS pipelines for protein–ligand and materials systems — setup, execution, trajectory analysis and rendering with MDAnalysis and custom tooling.
Electronic structure, applied.
B3LYP/6-311G(d,p) geometry optimisations and MO analysis in Gaussian 16 and ORCA — transition-state theory applied to real catalysts.
Search that cites its sources.
Hybrid dense+BM25 retrieval, FAISS indexing and LLM answer synthesis — published at TREC 2024 for verifiable tip-of-the-tongue search.
Instruments, not websites.
Browser-native tools scientists use daily: a 13-mode PDF studio, a multi-track DAW, chemistry calculators — all local-first, no uploads.
Molecules you can grab.
three.js and WebGL rendering with CPK palettes, instanced draw calls and measurement overlays — from ATP in the hero to full LAMMPS trajectories.
From idea to tested, shipped artifact.
Python, Java, C/C++ and modern web stacks with the testing and CI discipline to keep them alive — 500+ automated checks on this site alone.
From bond lengths to build pipelines — the details are the design.