Vaishnavi Jadhav

I'm Vaishnavi Jadhav, an MS Computer Science student at UIC graduating May 2027. I started out at IBM working on mainframe systems, then pivoted into AI/ML. I'm on a research and founder track now, building more than I take courses. Looking for somewhere innovative to work next.

The Record

Work Experience

1. Software Engineering Intern @ KNOWN (Cyclical Inc.)

   2026 — 2026

   Shipped production iOS and Android releases for a hormonal health startup using React, TypeScript, Capacitor, Supabase, SwiftUI, WidgetKit, Xcode, and Google Play Console. Built native Period and Daily Wisdom widgets with real cycle data, fixed Apple Sign-In and App Group data-sharing issues, and owned the mobile release pipeline through App Store Connect.

2. Graduate Research Assistant @ University of Illinois Chicago

   2026 — Present

   Researching NLP and causal AI for Alzheimer's biomarker discovery under Prof. Lu Cheng, with Prof. Natalie Parde as UIC co-advisor. Built and evaluated a guardrailed hybrid pipeline on 2,167 participants in the NACC autopsy cohort using FCI, literature retrieval, multi-model swap tests, bootstrap stability analysis, and NCSA Delta HPC. First-authored the 10-page paper submitted to IEEE Big Data 2026 on August 21, 2026; the work forms the basis of my MS thesis.

3. Application Developer @ IBM -- Bangalore, India

   2024 — 2025

   Maintained and migrated legacy enterprise applications on IBM mainframe systems. COBOL, z/OS, DB2 production pipelines. Zero-downtime, high-reliability systems where 99% uptime is considered a failure.

4. Frontend Developer Intern @ AHEN -- Pune, India

   2024 — 2024

   Built responsive web interfaces. Coordinated with designers and backend engineers. Final semester of undergrad.

1. 01

   Guardrailed Causal Discovery for Alzheimer's Biomarkers ↗

   2026

   First-author research on a guardrailed hybrid causal discovery pipeline over 2,167 participants in the NACC autopsy cohort. Combined FCI on rank-transformed clinical data with literature evidence, exogeneity constraints, swap tests, and bootstrap stability analysis. Found that LLM direction judgments flip with argument order while graph adjacencies remain stable and most orientations are sample-dependent. Submitted the 10-page paper "A Guardrailed Hybrid Pipeline for Causal Discovery of Alzheimer's Biomarkers" to IEEE Big Data 2026 on August 21, 2026.

   Python · pandas · NumPy · scikit-learn

2. 02

   Kindred

   AI-powered relationship intelligence engine that maps professional networks using graph-based knowledge representation (Neo4j). Tracks interactions, surfaces actionable follow-ups, and helps maintain meaningful connections. Won Best Project Award at TRAE/

   ReactJS · Node.js

3. 03

   Moral Machine: LLM vs Human Ethical Benchmarking ↗

   Systematic comparison of ethical decision-making between major LLMs and human responses from MIT's Moral Machine dataset. Trained classification models to measure moral divergence across different LLM architectures. Team of 5, course project.

   Python · pandas · NumPy · matplotlib

4. 04

   Delulify ↗

   Python tool that intercepts runtime errors and transforms them into emotionally supportive debugging hints. Because your stack trace should not make you cry. Open source, has a star and a fork.

   Python

5. 05

   UrbanScale ↗

   2026

   RL-based warehouse placement optimization for quick-commerce. Trained a MaskablePPO agent on real Chicago road topology (extracted via OSMnx) with a custom Gymnasium environment and multi-objective reward function. Agent outperformed greedy, random, and population-density baselines. Selected for NVIDIA Spark Hack (1 of 30 nationally). CS 594 Reinforcement Learning. Team of 3 -- designed the RL environment, reward model, and training pipeline.

   Python

6. 06

   LoopBack ↗

   2026

   Civic engagement app for reporting and tracking local issues in Chicago. Full-stack with Supabase backend, geolocation-based issue clustering, and deployed end-to-end. Built at DemonHacks (DePaul). Team of 6.

   Python · ReactJS · Cloud deployment

7. 07

   Pillar ↗

   2026

   Real-time news transparency platform that maps media ownership networks. Ground News shows political bias -- Pillar shows who is paying. Built a full Snowflake data warehouse with Cortex LLM inference (Llama 3.1-70b), FastAPI backend, RSS ingestion pipeline, and ElevenLabs-powered audio summaries. Top 3 finalist in Social Impact track at Uncommon Hacks 2026 (UChicago). Team of 2 -- built the data pipeline, LLM integration, and system architecture.

   Python · ReactJS · Tailwind CSS · RESTful API

8. 08

   EdgeRecall ↗

   Stateful AI assistant running entirely on Cloudflare's edge network. Uses Llama 3.3-70b via Workers AI for inference and Cloudflare KV for persistent conversational memory across sessions. Deployed and live. Built as a personal project exploring edge-native AI architectures.

   JavaScript · Tailwind CSS · Cloud deployment · RESTful API

9. 09

   Entropy-Based Selective Step Verification

   2026

   Research on making small LLMs reason faster without losing accuracy. Uses token-level entropy to identify uncertain reasoning steps, then selectively verifies only those steps using a Process Reward Model. Saves 51-57% compute. Tested on Qwen2.5-Math-1.5B, Llama 3.2 3B, and Gemma 3 4B across GSM8K and a custom options trading dataset. NeurIPS-format paper scored 9.8/10. Professor offered summer research continuation. CS 517 Socially Responsible AI.

   Python

10. 10

    Pathfinding Algorithm Visualizer ↗

    Interactive visualization of pathfinding algorithms (BFS, DFS, Dijkstra, A*) with real-time grid manipulation. Built to understand and demonstrate graph traversal concepts visually.

    Python

11. 11

    Redacted Lore ↗

    2026

    Collaborative storytelling platform where community-written stories are progressively uncovered through automatic redaction mechanics, inspired by SCP Foundation archives. Built with finite state machines for game logic, Redis for real-time state sync, and Devvit for Reddit-native deployment. Won the Reddit Daily Games Hackathon. Solo build, 48 hours.

    JavaScript · ReactJS

Case History

Coursework

1. CS 598 M.S. Thesis Research

   In progress, Fall 2026

   University of Illinois Chicago

   Thesis research on a guardrailed hybrid pipeline for causal discovery of Alzheimer's biomarkers using the NACC autopsy cohort. Primary advisor: Lu Cheng; UIC co-advisor: Natalie Parde.

2. CS 401 Computer Algorithms

   Done, Fall 2025

   University of Illinois Chicago

   Algorithm design, complexity analysis

3. CS 418 Introduction to Data Science

   Done, Fall 2025

   University of Illinois Chicago

   Data wrangling, visualization, ML pipelines

4. CS 594 Reinforcement Learning

   Done, Spring 2026

   University of Illinois Chicago

   MDPs, Monte Carlo, Q-learning, TD(lambda), function approximation. Project: UrbanScale (MaskablePPO warehouse optimization). Professor: Brian Ziebart

5. CS 491 Graduate Seminar

   Done, Spring 2026

   University of Illinois Chicago

   Attendance-based graduate seminar

6. CS 517 Socially Responsible AI

   Done, Spring 2026

   University of Illinois Chicago

   Fairness, bias, adversarial ML, conformal prediction. Project: Entropy-Based Selective Step Verification. Professor: Lu Cheng

7. CS 411 Artificial Intelligence I

   Done, Fall 2025

   University of Illinois Chicago

   Search, logic, probabilistic reasoning, ML fundamentals

8. CS 544 Machine Learning in Production

   In progress, Fall 2026

   University of Illinois Chicago

   Production ML systems, model quality, testing, MLOps, data infrastructure, and deployment reliability. Professor: Saeid Tizpaz-Niari

9. CS 554 Advanced Topics in Concurrent Computing Systems

   In progress, Fall 2026

   University of Illinois Chicago

   Advanced concurrent computing systems. Professor: William Mansky

• GitHubgithub.com/VaishJadhavVJ↗
• LinkedInlinkedin.com/in/vaishnavipjadhav↗
• Emailvaishnavipjadhav55@gmail.com↗