Harshal Desai
Software Engineer | MSc Computer Science @ UCD
Download ResumeAbout Me
A highly motivated Software Engineer with 3+ years of experience in scalable systems, currently pursuing MSc at UCD.
Education
- University College Dublin — MSc in Computer Science (2024–2025)
- NIT Jamshedpur — B.Tech (Hons.) in Computer Science & Engineering (2017–2021)
Industry Experience
Software Engineer — Lowe’s India, Bengaluru, India
June 2021 – July 2024
- Developed scalable microservices using Java and Spring Boot, supporting about 6k TPS (transactions per second).
- Designed and implemented REST APIs for efficient service communication.
- Gained experience working with Kafka for real-time event-driven processing.
- Managed CI/CD pipelines, including creating Docker images for consistent application deployment, and led testing efforts with Unit, Integration, and End-to-End Tests.
- Monitored and optimized performance metrics, focusing on P99 and P95 response times.
- Performed production monitoring using Grafana and Kibana, ensuring system reliability.
- Subject Matter Expert for Parcel Fulfillment applications, providing technical leadership.
- Promoted from Associate Software Engineer to Software Engineer within one year of joining as a fresher, in recognition of outstanding performance.
Technical Skills
- Programming Languages: Java, Scala, Python, SQL
- Tools/Frameworks: Spring Boot, Kafka, Git, Docker, Kubernetes, GCP, Jenkins, Numpy, Pandas, Scikit-learn, Matplotlib, Seaborn, Fastai, PyTorch, Keras
- Databases: PostgreSQL, Druid, Cassandra
- Others: Shell Scripting, Web Scraping using BeautifulSoup & Selenium, Groovy
Projects
Multimodal Image Retrieval System
Python, Fastai, sentence-transformers, nmslib | March 2021
- Built a text-to-image and image-to-image search system using the Flickr8k dataset.
- Used pretrained RoBERTa to generate 1024-d embeddings for text, and CNN for images.
- Mapped both into a common embedding space for semantic retrieval.
Sketch Image Clustering based on Semantic Similarity
Python, Fastai, SpaCy, Sklearn | June 2020
- Worked on Tu Berlin Sketch dataset (20,000 sketches, 250 categories).
- Fine-tuned ResNet34, achieving 78.90% accuracy — surpassing human performance.
- Clustered images using SpaCy vectors and pairwise cosine similarity.
Patience Game
Java, OOP | October 2024
- Console-based solitaire card game with a text-command system.
- Implemented card movements, lane rules, and game logic for foundation/stock handling.
- Features draw/move/quit commands for a fully playable CLI experience.
Skippy the Kangaroo Grid Simulation
Scala, OOP | September 2024
- Simulated a kangaroo using a Markov Chain model navigating a grid via dice rolls.
- Tracked Skippy's random journey and home arrival statistics.
Mineral Recognizer
Python, PyTorch, Fastai, AWS | May 2020
- Created a deep learning model (ResNet50) for mineral classification.
- Achieved 97.90% accuracy on validation data.
- Deployed serverless API on AWS Cloud using Python scripting.
Research
Multimodal Digital Media Analysis Lab, IIIT Delhi (April 2021 – Present)
Research Assistant | Guide: Dr. Rajiv Ratn Shah
- Created a multimodal web-scraped dataset with over 2M videos.
- Proposed efficient Webly Label training strategies using SSL and Multi-Label losses.
- Demonstrated robustness in downstream tasks with pretraining.
- Publication: NoisyActions2M (ACM MM Asia '21)
UE & HCI Lab, Department of Design, IIT Guwahati (June 2020 – August 2020)
Research Intern | Guide: Dr. Debayan Dhar
- Developed an automated system for evaluating uniqueness of design-based creative answers.
- Worked as a summer intern in the Usability Engineering and Human Computer Interaction Lab.
Achievements
- 99.36 percentile (AIR 7638) in JEE Mains 2017 amongst 1.2 million candidates.
- 94.43 percentile in JEE Advanced 2017 amongst 200 thousand candidates.
- Performed as a Young Tabla Artist on the ALL INDIA RADIO, Pune (2009–2013).
- Part of Taal Ninaad – recorded in GUINNESS BOOK OF WORLD RECORDS (2012).
- Top 1% in ML competitions (Analytics Vidhya, MachineHack)
- 9th place in Lowe’s Hackathon (3,820 teams)
Publications
-
Sharma, M., Patra, R. A., Desai, Harshal, Vyas, S., Rawat, Y., and Shah, R. R.
Noisyactions2m: A multimedia dataset for video understanding from noisy labels.
In ACM Multimedia Asia (New York, NY, USA, 2021), MMAsia ’21, Association for Computing Machinery.
[Read Paper]