04 — The surveyor
Route history
MS Computer Science at University of Illinois Chicago. Building tools where scale and responsibility share the same codebase. Currently based in Chicago. The short version: I like systems that are measurably correct and demonstrably fair. Résumé (PDF) ↗
Looking for now
Graduating May 2026 and open to full-time software engineering and applied ML / research roles — ideally where responsible AI, data systems, or geospatial work meet. Based in Chicago, open to relocation.
The throughline — fairness
I didn't come to fairness through theory — I took one course, pulled the thread, and couldn't put it down. Each project since has moved one layer closer to the systems that quietly decide things about people: first learning to measure bias, then to explain models, then to build the tools that catch them, and now to stress-test AI where the stakes are legal and civic.
Measure
Fall 2025
CS 516 · Responsible Data Science & Algorithmic Fairness, UIC
Toxicity in LLMs — a replication study
Replicated DExperts decoding-time detoxification on GPT-2 against RealToxicityPrompts, then broke it on adversarial ToxiGen. The lesson that hooked me: a method can score ~100% on explicit toxicity yet crack on implicit hate (98.5%) — and at a ~10× latency cost. You can’t fix what you can’t measure honestly.
Read the study ↗Explain
Spring 2026
Socially Responsible AI, UIC
Galaxy Morphology XAI
Benchmarked four post-hoc explainers (Grad-CAM, LIME, Integrated Gradients, GradientSHAP) across four CNN architectures on galaxy imagery, scored by faithfulness metrics. The finding: no explainer wins universally — rankings flip with architecture and dataset. Trust in a model has to be earned per-context, not assumed.
View project ↗Build
Spring 2026 · MS Project
Master's project, UIC
FairLint DL
Turned the research into a tool engineers actually touch: a deep-learning fairness debugger shipped as a VS Code extension, scanning models for demographic disparities in-editor via a two-phase gradient-ascent counterfactual search. Auditing time dropped 40% versus script-based workflows. Fairness belongs in the loop, not the postmortem.
View project ↗Investigate
2026 · now
Northwestern GAIN — Agentic AI Investigative Journalism Challenge
FairGuard
Reusable Agent Skills for investigative journalism, built with Claude Code and run over 1M+ federal lobbying records and congressional press releases. The anchor finding: the Artemis Group — founded by a former NASA Administrator — directs 39% of its filings at NASA with four former officials on staff, plus a ranked map of comparable revolving doors across 23 agencies.
View project ↗Benchmark
2026 · now
Advised by Prof. Saeid Tizpaz-Niari, UIC
TaxMR — robustness for tax LLMs
Low-income taxpayers increasingly ask AI for tax help, where a wrong or unstable answer costs a real refund. TaxMR uses metamorphic testing — swap a name, add an irrelevant sentence, reword the question, push back — and checks the answer doesn’t move. The thesis: neurosymbolic grounding, where only typed inputs reach an executor, is the most robust, because a name physically can’t change the result.
The pattern I keep chasing: AI is making more and more decisions about people who can’t see inside it. I want to keep building the instruments that hold it accountable — whether that’s an editor extension, an investigative agent, or a robustness benchmark.
Experience — 8 waypoints
Jun 2026 — Present
Chicago, IL
Machine Learning (Research) Engineer
RISC Lab · UIC
Metamorphic robustness benchmarking of tax-question-answering LLMs (TaxMR) in the Responsible, Informative, and Secure Computing Lab.
- Building TaxMR: 130 CPA-labeled VITA tax questions expanded to 2,080 metamorphic cases across four relations (demographic name swaps, irrelevant context, paraphrase, sycophantic pushback).
- Comparing robustness across base, RAG (IRS Pub 4491), and neurosymbolic (PAL) regimes on ~7 models; conditional violation rate with cluster-bootstrap CIs and exact McNemar tests.
- Advised by Prof. Saeid Tizpaz-Niari; Phase 1 pipeline validated end-to-end on Qwen 2.5 7B.
Aug 2025 — Present
Chicago, IL
Graduate Research Assistant
The Urban Transportation Center · UIC
Full-stack freight analytics toolkit and geospatial pipelines for 285+ Chicago-region municipalities.
- Engineered a Next.js + FastAPI freight toolkit visualizing crash metrics and traffic volumes across 285+ municipalities, deployed on UIC RHEL9 with Docker and Nginx.
- Architected Python pipelines processing 8.5M+ origin–destination pairs (300GB+) with chunked, atomic operations; containerized OpenTripPlanner to route across 2,926 zones.
- Built an OpenStreetMap Overpass extraction pipeline classifying 10,000+ US railyard polygons into intermodal, classification, and spur categories.
May 2025 — Aug 2025
Chicago, IL · Remote
Open Source Software Engineer
Google Summer of Code · OpenStreetMap
OpenLR-compatible temporary road closure database, API, and map UI for the OSM ecosystem.
- Built a temporary road closures database and RESTful API with PostgreSQL/PostGIS and FastAPI for real-time closure reporting in OSM navigation apps.
- Implemented OpenLR location referencing for cross-platform encoding across OSM-based navigation systems.
- Shipped a Next.js + Leaflet web app for community closure submissions; deployed on a dedicated OSM server with Docker, Nginx, auto-renewing SSL, and GitHub Actions CI/CD.
Feb 2025 — May 2025
Chicago, IL
Research Assistant
University of Illinois Chicago
Graph models of the Chicago road network for optimal congestion-zone analysis.
- Constructed a road network graph of Chicago with OSMnx and NetworkX, filtering to primary/secondary roads for computational efficiency.
- Engineered bounded cycle-detection algorithms to identify congestion zones and optimal camera placements within 3–5 miles of the city centroid.
Dec 2023 — Jul 2024
New York, NY · Remote
Software Engineer
SimPPL
Led a 5-person team; LLM hate-speech research and social network visualization tooling.
- Led a team of 5 building a project discovery platform for fellows and team leads across the organization.
- Co-authored a report with Prof. Deb Donig and the ADL on the impact of LLMs on hate-speech propagation in online forums.
- Built a coordination network graph visualizer used across YouTube, Truth Social, and BlueSky data.
Dec 2023 — May 2024
Boston, MA · Remote
Software & AI Engineer
Boston University
Virtual marketplace experiments on fake advertising and consumer behavior.
- Led a virtual marketplace experiment on Empirica.ly and React.js where producers choose between honest and deceptive ads while consumers select products.
- Worked with Prof. Marshall W. Van Alstyne and Dr. Swapneel Mehta in the Digital Information Research Lab.
Mar 2023 — Nov 2023
Remote
Software Engineer
SimPPL
Network analysis of YouTube video ecosystems via the YouTube Data API.
- Analyzed YouTube videos, comments, and channels; visualized video–user–comment relationships as network graphs.
- Shipped product pages for the company site, deployed on Vercel and Netlify with backend API testing and documentation.
Jun 2021 — Jul 2021
Mumbai, IN
Machine Learning Intern
Kaizen Future Tech
Stacked CNN/ResNet ensembles for natural disaster classification; Arduino IoT systems.
- Built CNN and ResNet image classifiers for natural disaster detection reaching 95% accuracy.
- Developed a stacked ensemble with XGBoost as meta-classifier to improve generalization.
Education
Aug 2024 — May 2026
University of Illinois Chicago
MS, Computer Science
GPA 3.7 / 4.0
- NLP
- Data Science
- Algorithmic Fairness
- Responsible AI
2021 — 2024
University of Mumbai · TSEC
BE, Information Technology
GPA 9.32 / 10
- DSA
- Machine Learning
- Computer Networks
- Data Mining
- Image Processing
Instruments
Languages
- Python
- TypeScript
- JavaScript
- SQL
- C/C++
- GLSL
ML & Data
- PyTorch
- scikit-learn
- PySpark
- Pandas
- NumPy
- LangChain / LangGraph
- Captum
- OpenCV
- Transformers
Web & APIs
- Next.js
- React
- FastAPI
- Node.js
- Tailwind CSS
- GSAP
- Three.js
- Leaflet / MapLibre
Geospatial
- PostGIS
- OSMnx
- NetworkX
- GeoPandas
- Shapely
- Valhalla
- OpenLR
- ArcGIS Pro
Infra
- PostgreSQL
- MongoDB
- Neo4j
- Redis
- Docker
- Nginx
- GitHub Actions
- Linux