MS.dev - Manpreet Singh

About Me

I'm a Computer Science student at Thapar Institute with a passion for pushing the boundaries of AI in biomedical applications. My work focuses on bridging the gap between cutting-edge deep learning and real-world biological challenges.

What drives me is the potential to accelerate scientific discovery through computational optimization. I've developed GPU-accelerated solutions that make previously intractable problems routine, from protein language model inference to medical image synthesis.

My approach combines rigorous academic research with practical engineering solutions. Whether it's optimizing CUDA kernels for bioinformatics algorithms or building physics-informed neural networks for biological modeling, I thrive on transforming computational bottlenecks into breakthroughs.

∞ Ideas Generated

42 Brainstorms Survived

7.5 Cups of Sanity Left

π Problems Solved (Almost)

Education

Bachelor of Engineering in Computer Science

Thapar Institute of Engineering and Technology

Patiala, India

Aug 2023 – Present (Expected 2027)

CGPA: 9.62/10.0

Research & Projects

High-performance AI systems for biomedical applications, from protein modeling to medical imaging.

BioCUDA_Triton: Protein LM Inference Engine

Custom CUDA + Triton backend for accelerating facebook/esm2_t6_8M_UR50D protein language models — dramatically cutting cost and latency for bio-ML workloads.

43× vs HuggingFace $1.1K/month saved 97.7% latency reduction

CUDA Triton Protein ML LLM Inference

Write-up

🧬

Speed

43×

Efficiency

97.7%

Savings

$1.1K

DeepMind Model Optimization: AlphaGenome & Enformer

Triton optimization of Google DeepMind's genomic attention models including AlphaGenome and Enformer for gene expression prediction.

5.05× faster AlphaGenome 1.82× faster Enformer Perfect accuracy preserved

Triton GPU Optimization Genomics Attention Models

GitHub

🔬

AlphaGenome

5.05×

Enformer

1.82×

Accuracy

100%

T1Converter: Multi-Modal MRI Synthesis

GAN-based pipeline for translating T1-weighted MRIs into T2, T1CE, and FLAIR modalities — reducing scan time and patient load in multiple sclerosis cases.

>95% SSIM 44min saved 70% cost reduction

PyTorch CycleGAN Medical Imaging

🏥

SSIM

95%

Time Saved

44min

Cost Cut

70%

Triton GPU Needleman-Wunsch

GPU rewrite of the classic Needleman-Wunsch algorithm using Triton, accelerating global sequence alignment in bioinformatics.

720× vs PyTorch 99K sequences/sec 1.7 GOPS compute

Triton Bioinformatics GPU Computing

GitHub

⚡

Speedup

720×

Throughput

99K/s

Compute

1.7G

ProteinBERT-Triton: Ultra-Fast Fill-Mask Inference

Custom Triton backend for accelerating ProteinBERT fill-mask predictions, turning a 5-year-old model into a modern speed demon while preserving biological accuracy.

3.6× avg speedup 100% accuracy $9.9K/year saved

Triton Protein ML Performance Engineering

🚀

Speedup

3.6×

Accuracy

100%

Savings

$9.9K

MolScribe-Triton: Molecular Image → SMILES

Triton-optimized backend for MolScribe, accelerating molecular image-to-SMILES translation with FlashAttention and mixed precision optimizations.

1.55× speedup 35.5% time reduction 100% SMILES match

Triton FlashAttention Chemoinformatics

Write-up

⚗️

Speed

1.55×

Time Cut

35.5%

Match

100%

PINN-Agent: LLM-Guided Multi-Agent System

Automated PDE discovery system with modular multi-agent PINN architecture and LLM controller for analyzing biological data and selecting PDE types.

99.97% R² synthetic 89.3% R² GBIF Automated PDE discovery

PyTorch Lightning PINNs LLMs GBIF Dataset

🤖

R² Synthetic

99.97%

R² GBIF

89.3%

Automation

Auto

CyberFlow: End-to-End MLOps Pipeline

Automated MLOps pipeline for detecting anomalies in the CIC-IDS2017 network dataset, covering traditional attack vectors and stealthy intrusions.

98.89% accuracy Automated retraining A/B testing

Airflow MLflow Docker Streamlit

GitHub

🛡️

Accuracy

98.89%

MLOps

Auto

Testing

A/B

Technical Arsenal

Comprehensive tech stack spanning GPU acceleration, AI/ML, and full-stack development.

🔡

Programming Languages

🧬

Machine Learning & AI

🤖

Deep Learning

⚡

GPU & Acceleration

📊

MLOps

🛠️

Tools & Platforms

Professional Experience

Research positions focused on AI optimization and biomedical applications.

Intern

CloudCosmos

North Carolina, USA

July 2025 - September 2025

Undergraduate Research Intern

Thapar Institute of Engineering and Technology

Patiala, India

July 2024 – Present

Leading research in AI bias, medical imaging, and physics-informed neural networks
Pretrained LLMs and VLMs from scratch and fine-tuned them for medical and robotics tasks
Built GAN-based MRI translation models (T1→T2, T1CE); developed PINN-based PDE solvers

Freelance Machine Learning Engineer

Stealth Startup

Remote

March 2025 – April 2025

Fine-tuned LLMs on federal immigration case documents to generate appeal reports
Improved IELTS prediction model: R² from 86% → 97%, MSE from 0.0208 → 0.0064
Built visa approval prediction model with 90% accuracy

Manpreet Singh

About Me

Education

Bachelor of Engineering in Computer Science

Research & Projects

BioCUDA_Triton: Protein LM Inference Engine

DeepMind Model Optimization: AlphaGenome & Enformer

T1Converter: Multi-Modal MRI Synthesis

Triton GPU Needleman-Wunsch

ProteinBERT-Triton: Ultra-Fast Fill-Mask Inference

MolScribe-Triton: Molecular Image → SMILES

PINN-Agent: LLM-Guided Multi-Agent System

CyberFlow: End-to-End MLOps Pipeline

Technical Arsenal

Programming Languages

Machine Learning & AI

Deep Learning

GPU & Acceleration

MLOps

Tools & Platforms

Professional Experience

Intern

Undergraduate Research Intern

Freelance Machine Learning Engineer

Get In Touch

Email

LinkedIn

GitHub