Surabhi Raghavan

I'm a

Experience

Software Engineer

Mobile Sensing + Health Institute, University of Pittsburgh, Pittsburgh, PA

Skills: Flutter · Dart · Java · Node.js · Python · Git · Docker · AWS (EC2, S3, Route53)

  • Led full-stack development of ROSA, a real-time remote monitoring app for 100+ chemotherapy patients, driving a 60% increase in early clinical interventions and a 35% reduction in unreported symptoms.
  • Deployed and maintained mobile app backends on AWS EC2 serving 200+ users/day.
  • Designed and deployed a Flutter frontend and Node.js backend to process 3,000+ patient surveys and 500K+ Fitbit data points monthly, maintaining 98% sync reliability.
  • Built a smart symptom-tracking system using branching PRO-CTCAE logic and automated alerts from patient-reported data and biometric thresholds, generating 124 clinician alerts/month.
  • Implemented a time-series anomaly detection engine using rolling baseline models to analyze HR, SpO2, and step count; applied statistical modeling and real-time pipelines to achieve 92% sensitivity in pilot evaluations.
  • Engineered feature pipelines for multimodal mobile-sensing data (surveys + wearable streams), generating daily/weekly aggregates such as trends, variability, and adherence to support downstream risk modeling for chemotherapy symptom worsening.
  • Built and validated supervised ML baselines (logistic regression) to predict next-day symptom escalation using passive signals (activity, sleep proxies) plus PRO-CTCAE history; reported AUC/F1 and performed patient-level cross-validation to avoid leakage.
  • Implemented production ML observability for real-world deployment, including drift detection on wearable signals (absences, steps, HR), alert-rate sanity checks, and automated retraining triggers to maintain stability over longitudinal patient use.
  • Operationalized privacy-preserving analytics by processing de-identified patient streams and enforcing secure data flows between mobile clients and cloud services to support remote monitoring research workflows.
  • Integrated MoSHI ecosystem tooling (e.g., sensor processing / dashboards) to improve QA on incoming wearable streams and accelerate ML iteration cycles for clinical-facing analytics.
October 2024 - Present

Software Engineer

Integra Micro Software Services, Bangalore, India

Skills: Figma · CSS · React · Node.js · REST APIs · UI/UX Design

  • Developed and deployed a custom real-time data validation framework to automate error detection and prevent data corruption, boosting platform reliability by 34% and reducing manual QA effort.
  • Optimized front-end performance by refactoring React components and reducing API overhead, achieving 25% faster load times and a smoother user experience.
February 2024 - May 2024

Software Engineer

IDS Next Business Solutions Pvt Ltd, Bangalore, India

Skills: Figma · CSS · React · Node.js · REST APIs · UI/UX Design

  • Drove major UX/UI overhauls via user research and wireframing, resulting in a 25% increase in app downloads and a 10% lift in platform revenue.
  • Architected and implemented mission-critical features in Flutter/Dart (dynamic search filters and real-time availability), which improved user trust and decreased booking cancellations by 25%.
November 2023 - February 2024

Researcher

Zeta Coding Innovative Solutions, Bangalore, India

Skills: Python · Machine Learning · Computer Vision · Flask · REST APIs

  • Developed a machine learning-based fraud detection system for images and videos, achieving 95% accuracy in detecting fraudulent content and strengthening platform security and trust.
  • Optimized the ML pipeline with automated preprocessing and evaluation, reducing iteration time by 32% and accelerating experiments.
August 2023 - October 2023

Projects

LLM Safety Classification

Multilingual Adversarial Prompt Classification for LLM Safety

My contribution: Built a multilingual safety classifier to detect adversarial and jailbreak prompts across English, Greek, French, and Arabic by extending HarmBench Behaviors dataset to 1,600 labeled prompts. Defined a 7-class harm taxonomy and fine-tuned XLM-RoBERTa-base with cross-lingual training, achieving 0.922 weighted-F1. Implemented robust preprocessing, stratified splits, and class-imbalance handling. Benchmarked Gemma 3 for zero-shot/few-shot safety classification and analyzed per-class failure modes and out-of-distribution robustness limitations.

Tech Stack: Python · Machine Learning · LLMs · XLM-RoBERTa · PyTorch · NLP · Fine-Tuning

FDA Analysis

Analyzing FDA Recall Trends Using Interactive Visualizations and Forecasting

My contribution: Led the end-to-end development of a robust analytical framework for processing and interpreting 96,000+ FDA recall records, uncovering significant temporal and categorical risk patterns that had direct implications for public health strategy. Designed and built interactive Altair dashboards that enabled stakeholders to dynamically explore recall severity and product-specific trends, shaping priorities for future risk monitoring and predictive modeling efforts. Played a critical role in bridging complex data analysis with strategic decision-making, establishing a foundation for scalable, ML-driven forecasting systems.

Tech Stack: Python · Altair · NumPy · Data Analysis · Pandas · Git

Cozy ToDo App

Cozy ToDo App

My contribution: Engineered the full-stack user authentication system, including secure login, registration, and email verification workflows. Integrated bcrypt for password hashing and configured sandbox email delivery through MailTrap to ensure safe and testable user communication. Built a comprehensive admin panel enabling privileged access to user account management, progress monitoring, and level editing capabilities. Ensured data integrity and access control across key workflows, contributing to both usability and security of the platform.

Tech Stack: Node.js · Express.js · Unity · EJS · nodemailer · MongoDB · bcrypt

Audio Anomaly Detection

Did you hear that? Introducing AADG: A Framework for Generating Benchmarking Data in Audio Anomaly Detection

My contribution: Designed and implemented a novel audio generation framework (AADG) leveraging large language models and text-to-audio pipelines to simulate realistic environmental scenarios for audio anomaly detection and localization. Led the expansion of training datasets beyond industrial settings, enabling model generalization to diverse, real-world audio domains such as telephony and multimedia. Built a modular, scalable system with rigorous validation workflows, significantly enhancing the robustness of anomaly detection models under out-of-distribution conditions.

Tech Stack: Python · Machine Learning

Flood Data Dashboard

Flood Data Dashboard

A flood-aware real estate assessment platform designed for Nepal, integrating geospatial and hydrological data to identify flood-prone zones and inform safer property decisions. The system enables users to browse, list, and evaluate properties with flood risk scores derived from climate, drainage, and urbanization data. Featuring a dual user model (owners and guests), it supports informed housing choices while promoting disaster resilience and sustainable urban development.

Tech Stack: Next.js · Tailwind CSS · Material UI (MUI) · MSSQL · Azure SQL · Node.js · JavaScript · HTML · CSS

Education

graduate

University of Pittsburgh

Master of Science
Information Science
August 2024 - Present

GPA: 3.94

Relevant Coursework:

  • Machine Learning
  • Interactive Data Science
  • Algorithm Design
  • Introduction to Natural Language Processing

Activities and societies:

SCIGSO, ANKUR: Indian Graduate Student Organisation

undergraduate

Visvesvaraya Technological University

Bachelor of Engineering, Honors
Computer Science

Relevant Coursework:

  • Artificial Intelligence & Machine Learning, Big Data Analysis, User Interface Design, Computer Graphics, Data Mining & Data Warehousing, Mobile Application Development, Compiler Design, Computer Networks, Unix Programming, Database Management Systems, Operating Systems, Object Oriented Concepts

Activities and societies:

Organiser: ANKURAM (Department Fest)

Skills

Programming Languages
Development Frameworks & Tools

Volunteer Works

Vice President

School of Computing and Information's Student Government
  • Represent student concerns and initiatives to school leadership, promoting transparency and communication.
  • Collaborate with the President and board members to lead meetings, plan events, and shape student government strategy.
  • Organize and support student-led initiatives including hackathons, tech talks, and community outreach programs.
  • Facilitate collaboration between student organizations, faculty, and university departments.
April 2025 - Present

General Secretary

ANKUR: Indian Graduate Student Association
  • Coordinate internal communication and maintain organizational records for continuity and transparency.
  • Assist in planning and executing cultural and networking events for the Indian graduate student community.
  • Liaise between board members, student body, and university offices to facilitate smooth operations.
  • Support onboarding of new students and foster an inclusive environment for all members.
  • Document meeting agendas and minutes while providing logistical and administrative support across events.
August 2024 - Present

Student Volunteer

Rotaract Club of BMS Yelahanka
  • Contributed to service-based initiatives aimed at community development and global citizenship.
  • Assisted in organizing awareness campaigns, donation drives, and local outreach programs.
  • Collaborated with team members and mentors to support leadership, education, and sustainability projects.
  • Promoted youth engagement and social responsibility through active participation in club events.
  • Represented student voices and values in planning community-centered activities.
June 2022 - June 2024