Building things that make life easier!

Hi, I am Uday, a software engineer based in Toronto, who is always exploring and open to new challenges.

Some visuals I vibe with

Engineering the Future

Roles, impact, and the systems I built — from production software to LLM research.

IBM • Markham, ON

Sep 2025 – Present

Led end-to-end development of a production-grade content creation platform for 2500+ authors and 10M+ learners, delivering a dual-mode WYSIWYG/Markdown editor built using Next.js, React, TypeScript, TipTap and Codemirror.
Shipped AI-assisted authoring features with document-level text-to-speech and content suggestions using GPT-4o-mini-TTS and OpenAI API, achieving 90% consistency with the source technical content and improved accessibility.
Built secure RESTful backend services with Express.js and OAuth2/JWT, integrating with a host authoring platform to enable draft, publish, and retrieval workflows with type-safe TypeScript contracts, reducing lab publishing time to less than 10 minutes.
Improved developer velocity with dockerized deployments and Playwright E2E tests, reducing regression bugs by 40%.

Google Developers Group (Glassbox LLMs) • Toronto, ON

Sep 2025 – Present

Built a transformer attention inspection pipeline on DistilBERT (6 layers, 12 heads) using Hugging Face Transformers and PyTorch to extract per-layer, per-head attention tensors for static and interactive visualizations via Matplotlib, Seaborn and BertViz.
Investigated long-context attention behaviour, analyzing inter-sentence attention to study how coherence and representations evolve with increasing sequence length, contributing reproducible analysis to an active area of interpretability research.
Developed a local, API-independent LLM evaluation pipeline by integrating DeepEval with a custom Ollama-backed model wrapper, enabling explainable LLM-as-a-judge evaluation using GEval metrics with thresholded pass/fail signals.

Blu Creative • Toronto, ON

May 2025 – Aug 2025

Developed an RAG chatbot using OpenAI API, LlamaIndex, LangChain, and SQLAlchemy, providing natural language query support over company data and RFPs, deployed across 50+ organizations.
Engineered AI-powered FastAPI endpoints with structured prompt engineering, chat history summarization, and query caching, enabling context-aware query handling, seamless Next.js integration, and a 30% improvement in response efficiency.
Built LLM screening to automate screening of 100+ daily documents into PostgreSQL, reducing manual processing time by 85%.

McMaster Aerial Robotics and Drone Team • Hamilton, ON

Jan 2025 – Present

Compared blob detection, hough circle transform, and connected-components methods for circular target detection, tuning parameters to avoid background false positives and achieving 90–95% accuracy on 100+ validation images.
Benchmarked detection accuracy and latency across approaches, selecting a blob-based pipeline that reduced inference time by 30% while maintaining stable detections under varied lighting conditions in images captured by a drone camera.

Learning Mode AI • Toronto, ON

Jan 2025 – Mar 2025

Built 3+ Go and Python based RESTful APIs to enable dynamic quiz generation, concise video summaries, and real-time Q&A by leveraging Redis and GPT-4o LLM integration, working in a docker based environment.
Developed 10+ UI features and resolved 15+ bugs, improving responsiveness by 30% using React and JavaScript.

Human Endeavour • Vaughan, ON

Jun 2024 – Oct 2024

Optimized a MySQL database for 15+ IoT sensors, boosting query performance by 40% via indexing and schema design.
Developed a C++ fall detection algorithm using an ESP32 micro-controller and accelerometer with 95% accuracy and implemented an automated email alert service via the Gmail SMTP server.
Developed a trilateration algorithm with Ultra-Wideband, AT Commands, and C++ for indoor tracking at 80% accuracy.