Data Scientist · AI Engineer · Researcher
Data scientist and AI/ML researcher with 4 peer-reviewed publications.
About
I'm a data scientist and AI enthusiast with a Master in Computer Science (Data Science specialization) from Seattle University and a Bachelor in Computer Engineering from the University of Pune. My research lives at the intersection of data quality and model behavior, the idea that better AI starts with better data, not just bigger models.
As part of a research team at Seattle University, I contributed to work on synthetic data generation for class-imbalanced medical datasets, published across IEEE Access, DaWaK, and DASFAA. Currently I'm building a personal AI memory system that captures how you think, not just what you know.
Experience
May 2025 — Present
Data Analyst
Boston Financial Advisory Group
Designed the data validation layer of an ML pipeline, building automated leakage checks to stop future settlement data from contaminating historical training sets, and resolved a sync-lag bug (via a SQL look-back window) responsible for 5% of apparent missing data, on top of the ETL automation that cut reporting time 40%.
Feb. 2023 — Apr. 2025
Research Data Scientist
Seattle University
ran extensive experimentation across synthetic-data generation and classification: built and benchmarked 5 SMOTE variants and 3 autoencoder architectures, designed a novel adaptive-control SMOTE algorithm, and trained/tuned multiple classical and transfer-learning classifiers on cloud GPUs — packaged into SDGnE, a reusable framework later adopted by other researchers, with results validated through a custom statistical evaluation pipeline and published across 4 peer-reviewed venues.
Jan. 2020 — Jan. 2022
Software Engineer
M.B.B. Consulting
Built cost-estimation models (Linear Regression, Random Forest) optimized on RMSE/MAE to improve pricing accuracy, and built an NLP pipeline (spaCy + regex) that cut PDF data-extraction time 80% (5 min → 1 min), saving 6.7 hours/week, alongside REST APIs and a 1M+ record ETL pipeline.
Projects
MindMirror
A personal AI system that learns how you think, not just what you know. Uses RAG and structured personal context to make every interaction feel like working with someone who has known you for months, without re-explaining yourself every time.
Personal projectAgent Orchestration Framework
An orchestration system that wraps Claude in a plan-act-reflect loop with typed tools, human-approval checkpoints for high-risk actions, and full audit logging, turning a stateless LLM into a durable, safely-supervised autonomous agent.
Published · Open sourceSDGnE Python Package
Open source Python package stemming from the SDGnE research project, lets users generate synthetic data from our designed algorithm for rare event and imbalanced classification tasks. Published research, usable tool.
View docs ↗ Personal projectTrail Recommendation AI Agent
An end-to-end AI agent that monitors calendar events, retrieves real-time weather data, and reasons across a personal trail database to deliver context-aware hiking recommendations, demonstrating full agent orchestration with tool use, memory, and multi-API reasoning.
Personal projectRAG Chatbot with Agentic Pipeline
A context-aware RAG chatbot built with LangChain and LLaMA3 fine-tuned with LoRA. Focused on production readiness — evaluating outputs critically, not just getting something that runs.
Research · PublishedWalkExplorer
A cloud-hosted multimodal AI tool on GCP using CLIP transformers and OpenStreetMap data to assess urban walkability. Benchmarked against human ratings with automated test validation, published at DASFAA 2026.
Read paper ↗ Learning projectTransformer LLM from Scratch
Trained a GPT model on the Shakespeare dataset using nanoGPT with character-level tokenization and AdamW. Achieved validation loss ~1.8, built to understand transformer architecture and ML math from first principles, not just use the API.
Research & Publications
IEEE Access · 2025
Synthetic Data Generation and Evaluation Techniques for Classifiers in Data Starved Medical Applications
DASFAA · 2026
Content-Based vs. Similarity-Based Deep Learning Approaches for Walkability Assessment
DaWaK · 2024
Incremental SMOTE with Control Coefficient for Classifiers in Data Starved Medical Applications
DASFAA · 2024
SDGnE: A Synthetic Data Generation and Evaluation System for Rare Event Prediction
Let's connect
I'm currently open to AI/ML engineering, applied research, and data science roles. If you're building something interesting or just want to talk AI, I'd love to hear from you.