Sahibnoor.

SahibnoorSingh Chahal

Data engineer turned ML-systems researcher. These days I write Triton and CUDA kernels and profile them down to the register.

Scroll

Now · since Aug 2026

Independent research · GPU kernel optimization

Hand-writing and benchmarking Triton/CUDA kernels — including scaled-dot-product attention — against PyTorch baselines on KernelBench ops. Custom softmax and matmul kernels reach ~1.9× over PyTorch eager and torch.compile at scale, profiled in Nsight Compute.

~1.9×speedup vs. PyTorch at scale

01Research

ML systems and GPU kernel optimization — in particular, using fine-grained hardware profiling signals like SM occupancy, memory bandwidth and register pressure to guide LLM-driven automated kernel generation, instead of black-box timing feedback.

  1. Aug 2026 – present

    Independent Researcher

    GPU Kernel Optimization Project

    • Writing and benchmarking hand-written Triton/CUDA kernels, including a scaled-dot-product-attention kernel, against PyTorch baselines using KernelBench ops.
    • Custom Triton kernels (softmax, matmul) benchmarked against PyTorch eager and torch.compile, achieving ~1.9× speedup at scale; profiled SM occupancy, memory bandwidth and register pressure with NVIDIA Nsight Compute.

  2. Jan 2020 – Jun 2020

    Research Intern

    Dept. of CSE, Bennett University · Advisor: Dr. Deepak Garg

    • Contributed to lab research on EEG-based emotion recognition: FFT feature extraction and genetic-programming classification.

  3. Jul 2017

    Summer Research Immersion

    Georgia Institute of Technology · Atlanta, USA

    • Developed image-processing and enhancement filters in Python; worked on embedded computing and object detection on the NAO humanoid robot.

02Selected projects

  • Deep learning
  • Bioinformatics
  • Evolutionary optimization

Protein sequence classification with CNNs + evolutionary search

Deep CNNs classifying unaligned amino-acid sequences into 17,000+ Pfam families, with architecture depth, dilation and kernel structure tuned over 50+ generations of evolutionary search. Evaluated on random and homology-clustered splits.

Result11–13% lower error than HMM and BLASTp baselines; >94% best-case accuracy.

  • Genetic programming
  • Neural architecture search

Evolving neural networks with genetic programming for breast-cancer classification

Encoded network architectures as GP trees and evolved them on the Wisconsin Breast Cancer dataset, with a novel hybrid crossover operator — constructive crossover plus fitness-gated hill-climbing — to keep diversity while converging faster.

ResultUp to 99% accuracy (70-30 split), beating constructive-crossover GONN.

  • Signal processing
  • Genetic programming

Emotion recognition from EEG signals

Full EEG workflow — Fourier-based frequency decomposition on 20-participant recordings — feeding GP classifiers with tuned fitness functions for four emotional states.

Result89% average accuracy, 6–10 points above SVM, Random Forest and NN baselines.

  • NLP
  • Multi-label classification

Categorizing online harassment language

Neural text classifier (embeddings, CNN and recurrent layers) for a binary label plus three harassment subtypes, on ~25k cleaned social-media posts with threshold tuning and class weighting.

Result8–12% higher multi-label F1 than statistical baselines on noisy Twitter data.

  • Generative models
  • Music

APRIL — A Program Reciting Intelligent Lyrics

LSTM models trained on 5,000+ lyric sequences and 500+ MIDI files, combining lyric and melody generation into one pipeline with temperature-controlled sampling.

ResultStylistically coherent lyrics and melodies, stable across sampling temperatures.

  • Gradient boosting
  • Analytics

Airbnb price prediction & market insights

Gradient-boosting pricing engine with end-to-end feature engineering in Python and Power BI dashboards turning predictions into decisions for hosts.

ResultRMSE 12.3; simulated 15% revenue gain through competitive pricing.

03Work

  1. Dec 2025 – Feb 2026

    Data Engineer (Contract)

    PRN Pharmaceuticals

    • Reconciled duplicate patient and practice records across 9 acquired-practice data sources into one consolidated NetSuite database.
    • Built a blocking-indexed fuzzy-matching pipeline (Python, rapidfuzz) with field-level normalization to score candidate matches at scale without brute-force pairwise comparison.
    • Built a synthetic data generator with controlled duplicate injection and realistic corruption to create labelled ground truth, then trained a Random Forest on similarity features to tune matching accuracy.

  2. Feb 2025 – Feb 2026

    Team Lead, Outreach & Fundraising

    Bright Mind Enrichment & Schooling

    • Led a cross-functional volunteer team of 40+; designed a donor-tracking system using engagement scoring to prioritize outreach.
    • Standardized data collection, storage and reporting workflows across a multi-channel fundraising operation.

  3. Dec 2020 – Jun 2023

    Data Engineer

    GagnaPro Solutions

    • Drove $7.5M+ in incremental revenue with churn-segmentation pipelines on 14M+ user records, enabling targeted retention campaigns.
    • Cut data refresh latency by 40% by building 20+ automated pipelines with Python, Airflow and AWS (EC2/Redshift).
    • Cut report generation time by 37% with scalable Snowflake data marts and optimized SQL; delivered 12+ Power BI dashboards with 50% faster load times.
    • Accelerated experimentation cycles by 30% through PySpark optimizations.

04Skills

GPU & parallel
TritonCUDANsight Compute / Systemstorch.compileKernel profilingRoofline analysis
Machine learning
Classification & regressionGradient boostingRandom forestsTime-series forecastingModel optimization
Data engineering
Entity resolutionFuzzy matchingAirflowDatabricksSnowflakeAWS Redshift
Languages
Python (pandas, numpy, sklearn, PySpark)SQLPSQLDAX

05Education

  • Sep 2023 – Dec 2024

    University of Massachusetts Amherst

    Isenberg School of Management

    M.S. Business Analytics · GPA 3.78

  • Aug 2016 – Jul 2020

    Bennett University

    India

    B.Tech. Computer Science, specialization in AI/ML · GPA 3.36

Certifications

  • MITx 6.86x: Machine Learning with Python
  • IBM Data Analyst
  • IBM Machine Learning Professional