I am an Advanced Data Science practitioner and Machine Learning researcher with demonstrated expertise bridging mathematical theory, scalable data pipelines, and commercial consulting impact.

My core focus lies across two high-impact verticals:

1. Healthcare & Life Sciences Analytics: Developing self-supervised Vision Transformers on 50,000+ clinical MRI records (96.3% precision, 97.0% AUC), cheminformatics reaction benchmarks using RDKit, and authoring the Global Medical AI Translation Observatory (GMATO) auditing regulatory approvals and healthcare data silos across 195 nations (validated by leaders at IIT Madras and Cambridge).
2. Quantitative Systems & Algorithmic Optimization: Formulating closed-form causal provenance graph algorithms that mathematically slash multi-agent attribution complexity from $O(2^K)$ to $O(K)$—eliminating reward variance traps and achieving out-of-sample Sharpe ratios of 1.15 on 15 years of National Stock Exchange (NSE) data (ACM SIGKDD 2026).

Healthcare Data Science, Regulatory Analytics, & Health Economics Across 195 Nations

• Problem: Over 90% of open-access clinical AI cohorts originate from the US and China, causing catastrophic demographic erasure in the Global South, while fragmented healthcare data silos drain institutions of $12.9M annually.
• Innovation: Engineered a 3-tier auditing observatory spanning scientific cohort provenance (PubMed, OpenAlex), national clearance registries (US FDA 510(k), EU CE-mark, India CDSCO), and demographic bias metrics across 195 nations.
• Deliverables: Full vector Apple Keynote & LaTeX Beamer presentations, 50+ page literature audit, and curated multi-national regulatory approval database.
• Validation: Reviewed and validated with Prof. Neil Lawrence (Cambridge) and Prof. Balaraman Ravindran (IIT Madras) at India AI Summit 2025.
• Tech Stack: Python, Pandas, PubMed API, OpenAlex, Keynote, LaTeX Beamer, Data Governance.
• 🔗 Repo: github.com/poppingpixel/GMATO

Deep Learning & Self-Supervised Vision Transformers on 50,000+ MRI Slices

• Problem: Conventional CNNs suffer from local inductive biases and high susceptibility to imaging artifacts in detecting intracranial neoplasms (gliomas, meningiomas, pituitary tumors).
• Innovation: Deployed a hybrid Vision Transformer (ViT-B/16) coupled with SimCLR Self-Supervised Contrastive Pretraining on axial brain MRI scans (T1, T2, FLAIR).
• Impact: Achieved 96.3% precision, 95.8% sensitivity, 96.5% specificity, and 0.970 ROC-AUC, decisively outperforming DenseNet-121 (92.5%) and ResNet-50 (91.2%) while reducing manual annotation requirements by 60%.
• Deliverables: Complete PyTorch implementation, diagnostic CLI, research manuscript under review at Springer Nature (European Biophysics Journal), and Apple Keynote / PowerPoint decks.
• Tech Stack: PyTorch, Torchvision, SimCLR, MONAI, Scikit-learn, NumPy, Keynote.
• 🔗 Repo: github.com/poppingpixel/ViT-Brain-Tumor-Detection

Provable Variance Reduction in Multi-Agent Financial Portfolio Construction

• Problem: When $N$ stochastic agents jointly produce a delayed financial reward, standard policy gradient estimators suffer an $O(N^2)$ variance explosion due to confounding pairwise covariances.
• Innovation: Formulated the Causal Provenance Ledger (CPL), a DAG that enforces a write-before-route protocol enabling deterministic counterfactual replay in linear time $O(K)$. Paired with Group Relative Policy Optimization (CPL-GRPO), proving gradient variance is bounded at $O(1)$.
• Empirical Results: Evaluated over 15 years of National Stock Exchange of India (NSE) data across the Nifty 50 universe, attaining an out-of-sample Sharpe ratio of 1.15 and Calmar ratio of 0.69.
• Publication: Accepted for Publication & Oral Presentation at ACM SIGKDD 2026 (FedMAS Workshop).
• Tech Stack: Python, PyTorch, LangGraph, Scipy, NetworkX, Pandas.
• 🔗 Repo: github.com/poppingpixel/QuantAlpha

Institutional Financial Strategy, Valuation Multiples, and Exit Dynamics

• Overview: Modeled capital deployment cycles, dry powder concentration, and valuation multiple compression across the Indian private equity and VC ecosystem.
• Deliverables: 17-slide vector presentation in LaTeX Beamer and institutional research report following Goldman Sachs Global Investment Research and McKinsey strategy frameworks.
• Tech Stack: LaTeX Beamer, TikZ, PGFPlots, Financial Modeling, Macroeconomic Analytics.
• 🔗 Repo: github.com/poppingpixel/india-venture-capital-analytics

Digital Signal Processing (DSP), Formant Tracking, and Cross-Lingual Acoustic Diagnostics

• Overview: End-to-end Python audio processing and sociolinguistic persona analysis pipeline leveraging Librosa and Praat (via praat-parselmouth).
• Capabilities: F0 pitch tracking, Burg formant algorithm ($F_1-F_4$), MFCCs, spectral centroids, and empirical acoustic distance analysis across General American, RP British, and Standard Indian English accents.
• Tech Stack: Python, Librosa, Parselmouth Praat, NumPy, SciPy, SoundFile, Matplotlib.
• 🔗 Repo: github.com/poppingpixel/voice-accent-acoustic-analysis

Computational Chemistry & RDKit Evaluation Harness for Foundation Models

• Overview: Benchmarks foundational LLMs on molecular graph representations, canonical SMILES, and ring-opening bond cleavage reactions across cyclic and heterocyclic scaffolds.
• Tech Stack: RDKit, Python, Chemical Graph Theory, LLM API Evaluation Harness.
• 🔗 Repo: github.com/poppingpixel/molecular-ai-cheminformatics

1. Deterministic Credit Assignment for Multi-Agent Portfolio Management via a Provenance DAG
   M. Rohit
   Accepted for Publication & Presentation at FedMAS 2026 Workshop (ACM SIGKDD 2026) • OpenReview Paper • GitHub Code

2. Sustainable Innovations in Brain Tumor Detection: Leveraging Vision Transformers for Enhanced Diagnostic Precision
   M. Rohit, S. Palanivel, K. Kannan
   Under Review at European Biophysics Journal (Springer Nature, submitted June 2026) • Manuscript PDF • GitHub Code

3. Global Medical AI Translation Observatory (GMATO): Quantifying Western Data Monopolies Across 195 Countries
   M. Rohit
   Academic Collaboration Proposal & Global Audit • Keynote Presentation Deck • Observatory Repo

I actively welcome discussions with research scientists, data science consulting leaders, and hiring teams:

• Personal Email: popingpixle@gmail.com
• University Email: rm9253@srmist.edu.in
• LinkedIn: linkedin.com/in/rohitphotography
• Research Portfolio: poppingpixel.github.io
• Direct Phone: +91 99108 23814