Hi, I'm Tobasum. I grew up moving a lot — six cities, eight schools before I turned sixteen — which is mostly a long way of saying I get comfortable in new places fast. These days I'm at Vanderbilt doing a BS/MS in Computer Science, with detours through Economics and Applied Math (currently out in the Bay for the summer).
I like problems where the math is real and the impact is too: AI agents, reinforcement learning, and ML for systems people actually use. Lately I've been nerd-sniped by training neural nets on fractal geometries — self-similar structures, hopefully more efficient architectures, that whole rabbit hole.
The thing I actually care about is building tools for communities that usually get left out of the design process. Most of the rest is downstream of that.
the quick version
BS/MS in CS @ Vanderbilt (+ Econ & Applied Math)
Venture Capital Intern @ G2C Ventures (prev. Venture Fellow @ IBM watsonx AI Labs)
Grew up across 6 cities in India, then UWC Mahindra (60+ nationalities, one roof)
Currently researching neural nets on fractal geometries
Technical due diligence on early-stage founders — poking at product feasibility and the engineering behind the pitch.
Venture Fellow · IBM watsonx AI Labs
Feb–Apr 2026
A self-directed growth agent for a live P2P API credit marketplace. Automated the GTM workflow end-to-end with LLM outreach — ~3x listings in month one.
SWE Intern · Vanderbilt Center for Transportation (CTOR)
Jan–Apr 2026
Agent-based sim of Nashville's traffic corridors (10,000+ patterns), plus RL for dynamic incentive allocation. Calibrated on 369,831 rideshare trips.
ML Intern · Advanced Learning Lab
Aug 2025 —
Multimodal ML predicting student metacognition on Carnegie Learning's MATHia (~88% accuracy, 50+ features).
NLP Intern · iTELL AI
Jan 2025 —
Domain-adaptive pre-training on ModernBERT / MPNet for comprehension scoring (~83%), plus PDF-to-JSON pipelines.
ML Research Intern · Category Lab (CATLAB)
Jun–Jul 2025
Vision-language models (CLIP, DNN/CNN) for visual cognition — 96% on ECOSET / ImageNet.
things I've built
Edge-AI earthquake search & rescue
won · AWS Autonomy
UAV–UGV coordination for post-quake SAR — thermal+RGB detection on Jetson, LiDAR SLAM in GPS-denied zones.
code →
Voice agent for blue-collar hiring
won · LaborUp
Redis-backed, MCP-powered voice agent handling 5K+ concurrent calls with live adversarial detection.
code →
Fractal pre-training for neural networks
working paper
Exploiting self-similar structure for better generalization. Pure math meets deep learning.
code →
Researcher-in-the-loop agent for quantum-classical code optimization
working paper
An AI agent that keeps a human in the loop while optimizing across the quantum-classical boundary.
code →
Mode-dependent linguistic variation in LLM-generated textbooks
working paper
A corpus-based comparative analysis of how LLMs write differently across modes.
Nashville traffic corridor sim
research
Agent-based model with R-tree indexing, calibrated on 369,831 trips to predict carpool adoption.
code →
Student metacognition prediction
research
Multimodal pipeline (XGBoost, NLP, time-series) predicting metacognition at ~88% on MATHia.
code →
Visual cognition models
research
CLIP / DNN vision-language models for perceptual expertise, GPU-accelerated across TF, PyTorch, MATLAB.
code →
say hi
I'm always up for research, hackathons, an AI-safety tangent, or a good book rec. Easiest to reach me by email — or find me on GitHub and LinkedIn.