I build the ML pipelines that turn protein models into design decisions: protein language models, structure prediction, candidate scoring and ranking, on HPC and cloud.

Open to full-time roles in protein design and AI drug discovery · open to relocation
sukhnansaran@gmail.com · LinkedIn

In silico antibody and protein design. Project details are under NDA.

• Built evaluation, filtering and ranking pipelines combining structure predictions with sequence- and structure-based scoring; on one antibody-design project this narrowed ~1,500 generated candidates to a top 30 for experimental testing.
• Integrated protein language model and structure-prediction pipelines into design workflows with computational biology and R&D teams.
• Optimised GPU structure-prediction runs with batched inference and multi-worker CPU parallelism.
• Deployed and orchestrated pipelines across AWS, SLURM and Nextflow, containerised with Docker.

Deep learning models deployed offline on Android, Raspberry Pi and Jetson Nano: 10–33 FPS, <350 ms end-to-end latency, model size cut from 10 MB to ~3 MB with quantization. Mentored 10+ interns.

Graphein — graph representations of protein structures · 4 merged PRs

• Fixed PyTorch Geometric conversion misaligning edge features with edge_index — #497
• Fixed hydrogen-bond detection in add_hydrogen_bond_interactions — #458
• Kept atom charges when saving PDB files — #498
• Fixed the extras flag on import messages — #459
• Under review: alternate-location conformer mixing — #515

BioPandas — pandas tools for molecular structure files

• Under review: Mol2 bond parsing — #170 · read_mol2 fix when ATOM ends the input — #171

Fine-tuning a diffusion model to redesign antibody CDR loops against a target epitope, with a held-out benchmark against the released weights. Write-up coming; code available on request.

alzheimers-rag · Python

Hybrid dense + BM25 retrieval with rank fusion, local LLM inference, cited answers and a hallucination guard. Evaluated on 44 labelled questions (hit@5, MRR); tested with pytest and GitHub Actions CI.

visa-rag-llm · FastAPI · Next.js · PostgreSQL · Redis · Celery

Self-hosted RAG system with an autonomous crawler, change detection, verified-answer path and a full pytest suite.

Computational biology — AlphaFold · ColabFold · ESM · ProteinMPNN · RFdiffusion · Biopython · PyMOL · Graphein · BioPandas

ML — Python · PyTorch · TensorFlow · Hugging Face Transformers & Diffusers · scikit-learn · XGBoost

HPC & MLOps — SLURM · Nextflow · AWS · Docker · batched GPU inference · ONNX · quantization

Backend — FastAPI · PostgreSQL · Redis · Celery · pytest

Working on protein design, antibody engineering or ML for drug discovery? I'd love to talk: sukhnansaran@gmail.com