About Me

Hiya, I'm Sirish Visweswar 👋, a machine-learning researcher and engineer with an M.S. in Computer Science from NYU. I'm a Research Engineer at the Neuroinformatics Lab with Prof. Erdem Varol, where I build self-supervised models for large-scale neural and imaging data.

My recent work has two threads. The first is localizing electrodes in the brain directly from the signals they record: geometry-aware self-supervised foundation models on raw Neuropixels LFP, published this year as two NeurIPS 2026 papers (a main-track paper and the LFP2Gene workshop paper). The second is high-speed voltage imaging on a 440 Hz two-photon microscope, in collaboration with Prof. Shy Shoham: I built its image reconstruction for the OJAMI paper (under review at Nature Methods), and a self-supervised denoiser that recovers individual action potentials from shot noise without ever seeing clean ground truth.

My interests sit at the intersection of deep learning, computer vision, and neuroscience, but I've also dabbled in reinforcement learning, game development, full-stack apps, blockchain, and NLP. In the long run, I'm looking to pursue a PhD or research-focused machine learning engineering roles in industry.

Outside of tech, I'm into creating art, which keeps me grounded and feeds into my problem-solving style: a mix of creativity, structure, and curiosity.

Updates

[Sep 2026] Two papers at NeurIPS 2026! "Geometry-Aware Self-Supervised Electrophysiology Representation Learning" in the main track and "LFP2Gene" at the BrainBodyFM workshop, both as co-first author. [Papers]

[May 2026] Completed my M.S. in Computer Science at NYU (GPA 3.85); continuing full-time as a Research Engineer at the NYU Neuroinformatics Lab.

[Jan 2026] Started as a Graduate Assistant for Neuroinformatics.

[Sep 2025] Started as a Graduate Assistant for Computer Vision.

[Jun 2025] Joined the NYU Neuroinformatics Lab with Prof. Erdem Varol.

[Sep 2024] Started my M.S. in Computer Science at NYU Tandon School of Engineering.

[Aug 2024] Graduated from BITS Pilani, Hyderabad with a B.E. in Electronics and Communication Engineering.

[Nov 2023] Published in IEEE Sensors Letters on smartphone deep learning for heart-attack detection from ECG.

[Jun 2023] Started my undergraduate thesis internship at the Speech & Language Lab, NTU Singapore.

Resume

Open PDF Download

Publications

Figure 1: pipeline from Neuropixels LFP chunks through audio foundation model backbones and geometry-aware contrastive objectives to 3D coordinate regression

Geometry-Aware Self-Supervised Electrophysiology Representation Learning | NeurIPS 2026 (Main Track) · Co-first author

2026 View Paper

Localizing recording electrodes in 3D directly from raw single-channel LFP, without histology, using probe geometry as a self-supervisory signal.

Geometry-aware contrastive objectives (CCFDistc, ChanDistc) constrain the embedding space to mirror physical channel layout along the probe
Benchmarks three audio SSL backbones (Wav2Vec 2.0, Whisper, Data2Vec) against supervised AnyNet, ViT, and spectral baselines
State of the art in brain-region classification and 3D coordinate regression across datasets, labs, species, and probe technologies; gains compound at larger data scales

Figure 1 of LFP2Gene: frozen wav2vec 2.0 LFP encoder with a decoder predicting Allen MERFISH gene expression around each channel

LFP2Gene: Recovering Local Molecular Composition from Local Field Potentials | NeurIPS 2026 BrainBodyFM Workshop · Co-first author

2026 View Paper

Self-supervised LFP representations retain signatures of the local molecular environment around the electrode.

Bridges IBL Neuropixels channels to Allen MERFISH atlases through the Common Coordinate Framework; a frozen encoder predicts expression of 1,122 genes
Recovers a good deal of local gene expression on 34 held-out subjects, more than any alternative under the same frozen-readout protocol
Shared structure is low-dimensional (PLS and CCA agree on five dimensions) and organised at the scale of brain regions

Denoised two-photon voltage imaging frame recorded with OJAMI

OJAMI: Electro-Optic Jitter Accelerated Multiphoton Imaging | Nature Methods (under review) · Co-author

2026

An add-on electro-optic scan module giving standard two-photon microscopes an order-of-magnitude speed-up at subcellular resolution.

Resonates along the mechanical scanner's slow axis, fast enough to record membrane voltage at 440 Hz
My contribution: image reconstruction, mapping each laser pulse's scattered sample onto a regular grid with an adaptive-bandwidth Gaussian kernel
S. Tucker, T. Reizis, E. Guralnik, D. Theagene, K. S. Guillory, V. S. Parupudi, D. Rinberg, F. St-Pierre, E. Varol & S. Shoham

A Smartphone-Enabled Deep Learning Approach for Myocardial Infarction Detection Using ECG Traces for IoT-Based Healthcare Applications | IEEE Sensors Letters, vol. 7, no. 11 · First author

November 2023 View Paper

Ensemble deep learning model for real-time myocardial infarction detection on mobile devices using ECG traces.

Developed majority voting ensemble with five quantized transfer learning architectures (FP16 and INT8 formats)
Achieved 99.34% accuracy using fivefold cross-validation scheme for mobile health imaging
Created portable smartphone-based healthcare system for real-time MI and heart abnormality detection

Projects

CineMatch – Group Movie Matching & Social Discovery Platform

Sept 2025 – Dec 2025 GitHub

Django + React movie discovery platform with personalized recommendations, community genre rooms, and real-time chat.

Deployed a Django + React movie discovery/tracking platform with personalized recommendations; built community genre rooms and private groups with synchronized swipes, real-time matches, and live chat via WebSockets + Redis reducing decision paralysis
Managed ZenHub backlog/velocity and shipped MVP/MLP features (indexed 50K+ titles, <300ms chat delivery, dashboards, catalog import/tracking, full-text search); deployed on AWS EB/EC2 with RDS and enforced quality via Travis CI (90%+ coverage)

Scalable Chest X-Ray Triage Pipeline with Spark & Med-GEMMA

Oct 2025 – Dec 2025 GitHub

Production-grade pipeline for automated radiology report generation with Kafka, Spark, and S3 data lake.

Engineered a production-style pipeline for automated radiology report generation with Kafka ingestion, Spark orchestration, and S3 data lake; including adaptive auto-tuning for GPU batch size and worker concurrency
Benchmarked stable scaling with 0.57–0.58 imgs/s throughput, 1.82s mean inference and 2.1s p95 latency; projected ~33K images/day capacity and 0.00007$ per report (>99% cost reduction)

Reinforcement Learning Environment for Scrabble

Mar 2025 – May 2025 GitHub

Custom OpenAI Gym-compliant Scrabble environment with multiple RL agents and comprehensive benchmarking.

Engineered environment with full rule fidelity, bag/rack simulation, and GADDAG-based prefix graph for legal move generation
Implemented Tabular Q-Learning, DQN, PPO, and Self-Play PPO agents with CNN encoders over (2 × 15 × 15) boards
Achieved 95% win rate vs. greedy/random agents using reward shaping, GAE, and top-50 move selection

PinToBeans

Jan 2025 – Apr 2025 GitHub

Full-stack social media platform with real-time features, content sharing, and comprehensive user engagement tools.

Developed Django REST API backend with PostgreSQL database, user authentication, and content management system
Implemented real-time messaging, post creation/editing, comment threads, and social interaction features
Built responsive frontend with modern UI/UX design patterns and seamless API integration

AD-FTD-CN EEG Classification using Multi-View Transformers

Jan 2025 – May 2025 GitHub

Deep learning pipeline using Multi-View Transformer for neurological condition classification from 19-channel EEG data.

Designed MVT with Graph Attention, cross-view fusion, and multiscale EEG embeddings integrating time, spectral, and spatial features
Achieved 98.55% within-subject and 64.95% cross-subject accuracy, outperforming SVM, MLP, and ADformer
Generated scalp heatmaps and attention maps for enhanced interpretability of lobe-specific frequency patterns

Det-IGEN: Detectable Watermarking in Image Generation

Sept 2024 – Dec 2024 GitHub

Invisible tree ring watermarking system for Diffusion Models to detect AI-generated content.

Implemented watermarking in the generation process by modifying initial noise vector in Fourier domain
Achieved perfect detection performance (AUC and TPR@1%FPR of 1.000) with negligible impact on image quality
Maintained high performance under adversarial conditions for robust AI content detection

Text Augmentation for NER on Noisy Language

2023 – 2024 Thesis

Entity-preserving text augmentation and LLM paraphrasing for named-entity recognition on noisy social-media language.

Built an augmentation pipeline using entity placeholders, BERTScore-guided candidate selection, and fuzzy retagging to prevent label drift
Extended the system with LLaMA 2 few-shot prompting, improving NER accuracy by approximately 8.6% over baseline
Limited entity loss to 6% while improving fluency by 24% on augmented text

Multi-Source Hate Speech Detection

Jan 2023 – May 2023 GitHub

Multi-class hate speech detection system using 11 Twitter datasets with advanced NLP techniques.

Developed system using 79k samples after preprocessing (URL/emoji removal, stopword removal, text normalization)
Implemented TF-IDF, GloVe, and BERT-based embeddings with ELECTRA, DistilBERT variants and various classifiers
Achieved ROC AUC 0.962, F1 Score 0.896 with ELECTRA + Logistic Regression using PCA optimization

Gaming Assets NFT Marketplace

Sept 2022 – Dec 2022 GitHub

Specialized gaming NFT marketplace on Polygon network for trading in-game assets as ERC-721 tokens.

Integrated marketplace enabling players to trade characters, skins, weapons with automated royalty distribution
Processed 500+ transactions with 2-second average confirmation time
Built responsive frontend with Web3.js integration and IPFS for decentralized metadata storage

Audio Visualiser Arduino

Oct 2021 – Nov 2021 Project Details

Audio spectrum visualizer with LED matrix that responds to music in real-time.

Created dynamic LED matrix display that visualizes audio frequency spectrum
Employed ESP32 microcontroller with Fast Fourier Transform (FFT) processing
Implemented programmable NeoPixels for responsive and colorful audio visualization

Games

Games developed as part of game design coursework and personal projects.

View Complete Game Portfolio

Zombieland

Survival horror game featuring zombie apocalypse scenarios with strategic gameplay elements.

Play Game

Outpost Epsilon

Space exploration game set in a remote outpost with mystery and adventure elements.

Play Game

Zero Vector

Gravity-shifting boss fight platformer game featuring dynamic physics-based gameplay mechanics.

Play Game

Penguin Adventure

Charming adventure game featuring a penguin protagonist in arctic environments with puzzle elements.

Play Game

Teaching

Graduate Courses (NYU)

100+ graduate students across both courses · 8 Kaggle competitions curated

CS-GY 6643: Computer Vision

Graduate Assistant · Fall 2025

Graduate course from classical image processing to modern deep learning for vision.

Ran hands-on code labs with live coding and guided notebooks
One-on-one project guidance in 6 hours of weekly office hours
Graded assignments, projects, and exams with detailed feedback
Built assignments and organized a class-wide constellation-detection Kaggle competition

CS-GY 9223: Neuroinformatics

Graduate Assistant · Spring 2026

Graduate course at the intersection of neuroscience, machine learning, and large-scale neural data.

Ran live coding sessions, office hours, and research-project mentoring
Authored and graded assignments and exams across core and advanced neuroscience topics
Curated Kaggle competitions end to end: datasets, baselines, metrics, and submission feedback

Teaching Resources

Materials I built for the courses

Notebooks: computer vision algorithms, feature detection, and deep learning with hands-on implementations
Assessments: programming assignments and projects implementing CV algorithms on real-world images
Kaggle: a class-wide constellation-detection competition for end-to-end model development and evaluation

Guest Lectures & Workshops

Introduction to Robotics Workshop

BITS Pilani 2022-2023

Mentored workshop guiding students through Arduino-based line follower bot development, covering sensor integration, motor control, and algorithm implementation.

60+ students Arduino projects

View Materials

Techtainment Robotics Project

BITS Pilani 2022-2023

Guided a team of 5 students in developing an innovative robotics headgear project, providing mentorship in hardware integration, programming, and project management.

5 team members Headgear robotics

View Project

About Me

Updates

Resume

Publications

Geometry-Aware Self-Supervised Electrophysiology Representation Learning | NeurIPS 2026 (Main Track) · Co-first author

LFP2Gene: Recovering Local Molecular Composition from Local Field Potentials | NeurIPS 2026 BrainBodyFM Workshop · Co-first author

OJAMI: Electro-Optic Jitter Accelerated Multiphoton Imaging | Nature Methods (under review) · Co-author

A Smartphone-Enabled Deep Learning Approach for Myocardial Infarction Detection Using ECG Traces for IoT-Based Healthcare Applications | IEEE Sensors Letters, vol. 7, no. 11 · First author

Projects

Games

Teaching

Graduate Courses (NYU)

Teaching Resources

Guest Lectures & Workshops

Tags