01
6+ Top-Tier Publications
Published in Nature Computational Materials, JACS, ACS Cent. Sci., and Nano Letters as first or equal-contribution author.
Open to AI/ML Scientist Roles · Drug Discovery
Hi, I'm Sourav.
AI/ML Scientist · Computational Chemist · Drug Discovery
Computational scientist with 10+ years bridging quantum chemistry and deep learning to accelerate drug discovery. Currently at Aganitha Cognitive Solutions — building end-to-end pipelines for crystal structure prediction, conformer generation, pKa estimation, and ADMET property calculations. Published in Nature Computational Materials, JACS, and Nano Letters.
Milestones
Key
Key
Achievements
A proven track record of shipping production pipelines, autonomous agents, and highly-cited research in computational drug discovery.
02
6+ Production Pipelines in Igniva™
Built CrystalCleanPro, CrystalLatticeAI, MolConSUL Pro, SitepKa, QM-pKa, and ADMET platforms — shipped as part of Aganitha's commercial Agentic AI product.
03
AI Agent for Lead Optimization
Autonomous medicinal chemistry agent using Claude AI with iterative propose-score-compare loops for drug lead optimization.
04
End-to-End Drug Discovery Platform
Full-stack triage app (React, FastAPI, RDKit) with real-time ADMET, QED, and PAINS screening deployed on Vercel.
Computational Stack
AI/ML for Drug Discovery
- Graph Neural Networks
- PyTorch / TensorFlow
- Transformers
- ADMET Prediction
- AI Agents / LangChain
- Bayesian Optimization
Computational Chemistry
- CP2K / ORCA / PySCF
- DFT / SAPT
- ML Force Fields
- Molecular Docking
- pKa Prediction
Cheminformatics
- RDKit / ASE
- SMILES / SMARTS
- Conformer Generation
- Crystal Structure Pred.
- ADMET Workflows
MLOps & Languages
- Python (Expert)
- Docker / Azure
- CI/CD / MLflow
- NumPy / Pandas
- Fortran / Bash
Projects & Research
Scientist · Aganitha Cognitive Solutions · Jan 2024 – Present
Production AI/ML Pipelines for Pharmaceutical R&D
PyTorch
GNNs
RDKit
CP2K
Docker
Azure
Crystal Structure Prediction
GNN pipeline for rapid polymorph screening of drug-like molecules — predicting stable crystal forms critical to formulation.
Conformer Generation
Environment-aware AI platform generating realistic 3D conformers accounting for solvent and protein-pocket context.
pKa Estimation
Dual-model approach — QM-based and GNN-based — for accurate pKa prediction under non-aqueous conditions.
ADMET Prediction
Physics-based QM platform predicting solubility, logP, and solvation energy for high-accuracy ADMET profiling.
AI Agents
Claude AI · HuggingFace
Lead Optimization Agent
AI-powered medicinal chemistry sandbox using Claude to iteratively propose and score structural modifications. Autonomous loop: propose → score → compare → iterate with full property trajectory tracking.
- BBB, CNS MPO, QED, and flexibility tracked per attempt
- Highlighted 2D structures showing structural changes
- Local run persistence — resume without extra LLM cost
GNN · R²=0.90
Graph Neural Networks
Solubility Dual-Graph GNN
Dual-graph interaction GNN predicting molecular solubility (logS) from solute-solvent SMILES pairs. Bidirectional cross-attention mechanism trained on 100K+ BigSolDB 2.0 pairs achieves R² = 0.90, RMSE = 0.388.
- Separate MPNN encoders for solute & solvent graphs
- Bidirectional cross-attention for interaction modeling
- 100,983 solute-solvent pairs from BigSolDB 2.0
GNN · RL · MLflow
Binding Affinity · REINFORCE
GNN Bind Optimizer
Heterogeneous GNN for protein-ligand binding affinity prediction paired with a REINFORCE-based molecular generator. End-to-end pipeline with SQL Server persistence, MLflow experiment tracking, and an interactive Streamlit UI for pocket-aware ligand design.
- HGTConv heterogeneous GNN — MTL & STL training modes
- REINFORCE RL loop for pocket-conditioned molecule generation
- Docker Compose full-stack: SQL Server + MLflow + Streamlit
Bayesian Opt · GP
Tablet Formulation · BoTorch
Formulation Bayesian Optimization
Model-based DoE pipeline replacing OFAT/grid screening with uncertainty-aware sequential experimentation for pharmaceutical tablet formulation. GP surrogate models maximize Q45 (% drug dissolved at 45 min) under hard mass-balance constraints across a 5-excipient design space (HPMC, MCC, CCS, MgSt, PVP K30).
- Single-objective BO: GP + EI/LogEI to maximize Q45 with fewer wet-lab runs
- Multi-objective BO: qNEHVI Pareto front over Q45, hardness & friability
- Streamlit dashboard: design-space exploration, experiment logging & convergence comparison
RAG · Structure Search
InChIKey · BM25 · RDKit
ChemKeyRAG — Structure-Keyed Literature RAG
Retrieval over chemistry papers keyed on structure, not on the name an author happened to type. Every chemical name is resolved to an InChIKey skeleton at ingest, so one molecule retrieves every passage about it — paracetamol, acetaminophen and 4-hydroxyacetanilide collapse onto one key. No vector database, no embeddings, no trained NER.
- Entity linking: lexical candidates → RDKit-verified lexicon → PubChem, cached
- Measured, not asserted: a systematic-name query returns 0 text hits vs 115 by structure
- Molecule-pinned cited chat — refuses rather than substituting another compound’s data
Featured
Full-Stack · Live Demo
Drug Discovery Triage Platform
Full-stack drug discovery triage application with real-time ADMET property predictions, QED scoring, PAINS alerts, and interactive 2D molecular visualization for rapid compound screening.
- Real-time ADMET: lipophilicity, solubility, permeability
- QED drug-likeness scoring & PAINS substructure alerts
- Dockerized deployment with CI/CD pipeline
Recent News
Preprint
ChemRxiv
Preprint posted: SitepKa
"SitepKa: Site-Specific Acid and Base pKa Prediction Across Multiple Industrial Solvents" — with Antarip Halder and Chanukya Nanduru. Dual-graph cross-attention GNN predicting site-specific pKa across water and 34 industrial organic solvents.
May 2026
New Project
Launched: Formulation Bayesian Optimization
Released model-based DoE pipeline for pharmaceutical tablet formulation. GP surrogate + EI/qNEHVI acquisition maximizes dissolution (Q45) and multi-objective trade-offs under hard excipient constraints — replacing costly OFAT screening.
2025
Publication
Paper published in ACS Central Science
"The spin phonon relaxation of single molecules magnet in the presence of strong exchange coupling" — first-principles study linking phonon coupling to spin relaxation rates in molecular magnets.
Jan 2024
Role
Joined Aganitha Cognitive Solutions as Scientist
Building production AI/ML pipelines for pharmaceutical R&D — crystal structure prediction, conformer generation, pKa estimation, and ADMET profiling within Aganitha's commercial Igniva™ platform.
2023
Publication
Paper published in npj Computational Materials
"Spin-phonon decoherence in solid-state paramagnetic defects from first principles" — ab-initio framework for computing spin coherence times in qubit-relevant defect systems.
Selected Publications
Published in Nature Computational Materials · JACS · Nano Letters · J. Phys. Chem. Lett. · Preprint on ChemRxiv
Preprint
ChemRxiv
SitepKa: Site-Specific Acid and Base pKa Prediction Across Multiple Industrial Solvents
Sourav Mondal, Antarip Halder and Chanukya Nanduru
2025
ACS Cent. Sci.
The spin phonon relaxation of single molecules magnet in the presence of strong exchange coupling
Sourav Mondal, Julia Netz et al.
2023
npj Comput. Mat.
Spin-phonon decoherence in solid-state paramagnetic defects from first principles
Sourav Mondal and A. Lunghi
2022
JACS
Unravelling the contributions to spin-lattice relaxation in Kramers single-molecule magnets
Sourav Mondal and A. Lunghi — J. Am. Chem. Soc., 50, 22965
2022
J. Phys. Chem. Lett.
Identification and Manipulation of Defects in Black Phosphorus
R. Harsh, Sourav Mondal*, D. Sharma et al. — 13, 6276 *Equal contribution
2020
Nano Letters
Direct Observation of the Reduction of a Molecule on a Nitrogen pair in Doped Graphene
M. Bouatou, Sourav Mondal*, C. Chacon et al. — 20, 6908 *Equal contribution
2019
J. Phys. Chem. Lett.
Importance of Epitaxial Strain at a Spin-Crossover Molecule Metal Interface
C. Fourmental, Sourav Mondal*, R. Banerjee et al. — 10, 4103 *Equal contribution
Experience & Education
Current
Jan 2024 – Present
Scientist
Aganitha Cognitive Solutions · Hyderabad
- Developing CSP pipelines with Graph Neural Networks for drug polymorph prediction
- Building AI models for conformer generation, pKa prediction, and ADMET properties
- Creating thermodynamic workflows: solubility, logP, and solvation energy
- Managing projects, client communications, and proposal writing
Previous
Aug 2023 – Dec 2023
Computational Scientist & Deep Learning Engineer
QpiVolta Technologies · Bangalore
- Developed machine learning force fields for solid-state electrolytes
- Modeled thermodynamic properties using molecular dynamics
- Led contract research on complex biological systems simulation
Postdoc
Mar 2021 – Jul 2023
Postdoctoral Researcher
Trinity College Dublin · Ireland
- Built neural network models for quantum property prediction
- Developed computational methods for spin dynamics simulation
- Published in Nature Computational Materials and JACS
- Combined ab-initio methods with machine learning approaches
PhD
Jan 2015 – Feb 2021
PhD in Computational Material Science
JNCASR · Bangalore
- Thesis: Tailoring Properties of 2D Systems via Molecular Adsorption and Defect Engineering
- DFT study of electronic and magnetic properties of 2D materials
- Published in Nano Letters and J. Phys. Chem. Letters
- Received CEFIPRA grant for research visits to Paris
M.Sc
2012 – 2014
M.Sc in Chemistry
IIT Guwahati
Qualified NET CSIR-UGC Junior/Senior Research Fellowship