Software engineer building full-stack and AI-integrated systems — see my portfolio for detailed project write-ups.

Student Government Association (SGA) — Boston, MA
Software Engineer | Jan 2026 – Present

• Developing a production content management system serving 5,000+ students, enabling non-technical editors to publish independently and cutting publish turnaround time by 10%
• Architecting version history, rollback capability, role-based access control, and soft-delete archiving on a Next.js and Prisma stack to support safe, auditable content operations
• Engineering Prisma-validated API endpoints that enforce schema integrity on every mutation over a PostgreSQL backend
• Shipping features in Agile/Scrum sprints via Linear, coordinating scope directly with team leads and delivering through peer-reviewed pull requests

Disrupt: The FinTech Initiative — Boston, MA
Quantitative Analyst | Jan 2026 – Present

• Building scalable Python backtesting pipelines to evaluate systematic trading signals across 10M+ historical market data points
• Developing and deploying pairs-trading, mean-reversion, and momentum strategies against $50K of simulated capital
• Translating equity-market research on statistical arbitrage, factor investing, and market microstructure into testable, production-ready trading signals

Khoury College of Computer Sciences — Boston, MA
Teaching Assistant, Program Design and Implementation II | May 2026 – Jun 2026

• Led 30+ weekly debugging sessions in Java for a cohort of 100+ students, coaching object-oriented design, data structure implementation, and control-flow debugging strategies
• Taught AI-assisted development workflows, integrating Claude and GitHub Copilot into the curriculum to teach effective prompt engineering alongside traditional debugging
• Graded assignments against detailed rubrics, delivering written feedback on SOLID principles, code architecture, and correctness to reinforce best practices

Rainfall Learning — Boston, MA
Software Lead | Nov 2025 – Jun 2026

• Leading front-end development on a real-time collaborative IDE alongside 15+ engineers, reducing edit conflicts to under 2% through CRDT-based synchronization with Yjs
• Building a scalable React and TypeScript UI layer optimized for concurrent multi-user editing, improving render performance by 20%
• Dockerizing build pipelines to standardize local and production environments, cutting environment-related failures by 15%
• Engineering component and integration test suites with Vitest and React Testing Library, catching UI regressions pre-merge and improving overall frontend reliability

Multimodal Machine Learning for Parkinson's Disease Detection — Northeastern University
Undergraduate Researcher - Advised by Prof. Sarita Singh | 2026 – Present

• Developing the machine learning layer for an edge-computing wearable Parkinson's monitoring system, extending a published IEEE framework (2026 IEEE World AI IoT Congress) that captures resting tremor and bradykinesia from finger-mounted IMU and flex sensors on a Raspberry Pi edge device
• Designing a subject-level, leakage-resistant evaluation protocol — three-way train/validation/locked-test split, per-subject cross-validation, and pre-registered significance testing — to produce honestly benchmarked results and avoid the inflated per-recording metrics and selection bias common in digital-biomarker research
• Conducting a critical review of 10+ prior studies to isolate an open research gap: no existing work fuses multiple digital biomarkers (voice, finger-tapping, and gait) from the same subjects while comparing several fusion strategies (early, late, gated, and attention) under per-subject evaluation with statistical testing
• Building single-modality baselines across classical and deep algorithms (logistic regression, SVM, gradient boosting, 1D-CNN/CNN-LSTM) as a benchmark for whether multimodal fusion yields a statistically significant gain over the strongest single modality, with a per-subject fusion pipeline (per-modality encoders → fusion → calibrated MDS-UPDRS-aligned output) designed for low-cost INT8/TFLite edge deployment

AI Coding Tools in Programming Education — Northeastern University
Undergraduate Researcher - Advised by Prof. Sarita Singh | 2026 – Present

• Preparing two work-in-progress poster submissions for the ACM Technical Symposium on Computer Science Education (SIGCSE TS 2027) on integrating AI coding tools into introductory programming courses without eroding students' foundational skills
• Leading the drafting of the 2-page ACM extended abstracts in LaTeX (acmart/sigconf), and supporting the design of a differential "human-baseline vs. AI" evaluation protocol in which students solve a task unaided, then with an AI tool, then perform a structured gap analysis on a shared rubric covering correctness, requirement coverage, edge-case handling, and security
• Comparing four AI coding tools — GitHub Copilot, Cursor, ChatGPT/Codex, and Claude Code — across four programming paradigms (Python, object-oriented Java, Prolog, TypeScript/JavaScript) and task types including generation, debugging, testing, and refactoring
• Applying an integrated TPACK–SAMR–TAM–Bloom framework linking instructional design, tool-integration depth, student acceptance, and learning outcomes, alongside a phased model for sequencing AI use against competence gates

NU Dining

• Developed a full-stack web application with a team of 5, integrating the DineOnCampus API via automated serverless cron jobs to serve real-time daily menus from Northeastern University's three dining halls
• Designed and implemented a responsive, accessible UI using React, Vite, and TailwindCSS, translating Figma prototypes into production-ready components with dietary filtering and nutrition breakdowns
• Architected a Supabase PostgreSQL backend with row-level security, managing a relational schema across locations, periods, stations, menu items, and nutrients to support efficient querying at scale
• Built a user authentication system with Supabase Auth enabling personalized features including a calorie tracker and meal voting system, while maintaining full menu access for unauthenticated users
• Deployed the application on Vercel with serverless API routes, environment-scoped secrets, and a daily cron pipeline that automatically scrapes and upserts menu data to keep content current without manual intervention

Reinforcement Learning for Derivative Hedging

• Framed dynamic option hedging as a continuous-action MDP and trained PPO and SAC agents to hedge a short European call position, benchmarked against Black-Scholes delta hedging across four market scenarios: base, high transaction cost, volatility mismatch, and regime switching
• Built a custom OpenAI Gym environment replaying 5 years of real SPY daily price data across 1,204 overlapping 30-day windows, exposing agents to the 2020 COVID crash, 2022 rate shock, and 2023–24 bull market within a single training distribution
• Designed a 6-feature normalized observation space (normalized spot price, time-to-expiry, delta, gamma exposure, current hedge position, and log-moneyness) with an asymmetric reward penalizing downside P&L variance and a terminal settlement penalty, making the agent explicitly risk-averse rather than variance-neutral
• Evaluated 5 strategies (PPO, SAC, delta hedge, no hedge, random) across all scenarios reporting 8 metrics (Sharpe, VaR 95%, CVaR 95%, mean/std P&L, max loss, % loss episodes, avg transaction cost); PPO improved Sharpe over delta by ~33% in the high-TC regime and ~37% in the volatility mismatch regime
• Built a 5-page Streamlit dashboard covering live episode animation (agent vs delta hedge step-by-step), real-time training with live learning curves for both agents, full evaluation results, a Monte Carlo scenario lab, and a live SPY options chain with implied vol surface

📧 patel.s15@northeastern.edu