Welcome to my portfolio! I'm a software developer with a passion for applied machine learning. My professional experience combines robust backend development, microservices, NLP, and computer vision. I hold a strong academic background and believe great software addresses both technical excellence and real-world impact. Explore my projects and let's connect!
Full-stack software engineer embedded in the Pricing and FX Payments space, working across the entire stack, from Java backend services handling millions of low-latency transactions daily, to React-based merchant-facing workflows. I took ownership of complex, cross-functional initiatives rather than just executing tickets: I led the end-to-end expansion of PayPal's Merchant Hub into Middle Eastern and African markets, aligning engineering and business stakeholders to unlock $17M in new annual revenue. Alongside this, I am also working on AI initiatives to streamline triaging and debugging flows. I designed and shipped a multi-agent orchestration system for FX transaction analysis, cutting live support turnaround time by 75%. It was one of the first production deployments for Payments org using Google ADK, CrewAI, MCP, and RAG, and required both technical depth and the ability to drive adoption across teams.
Technologies Used: Java, SpringBoot, React, Next.js, CrewAI, RAG, MCP, Python, Harness.
Built production-grade AI agentic onboarding assistant adopted by 100+ new hires. I owned the end-to-end Q&A workflow, delivering contextual, personalized guidance that accelerated time-to-productivity by 60%. Beyond the pipeline, I designed the prompt engineering layer from the ground up, experimenting with zero-shot and few-shot in-context learning across both open- and closed-source LLMs to generate accurate, code-centric responses. On the infrastructure side, I deployed a FastAPI-based ML inference service, tuning retrieval performance against evaluation metrics like Recall@K and MRR to ensure the assistant held up at scale.
Technologies Used: PyTorch, GPT, CodeLlama, Python, Flask, Faiss, Elasticsearch, FastAPI, CodeT5.
Improved performance of the Cobalt, Intel's graphics simulation module, effectively boosting CPU frequencies while fortifying against potential future regressions. Identified and improved suboptimal algorithms, harnessing tools such as V-Tune to pinpoint and address bottlenecks leading to a 300% surge in CPU frequencies. I also devised scripts for generating visualization plots and scrutinizing upcoming commits, enabling vigilant monitoring to detect regression-inducing changes.
Technologies Used: C++, Python, VTune, Visual Studio.
Worked for the LiveDesign team, a platform for drug discovery and material science, where I managed a small engineering team, led projects, and optimized features. As the engineering lead, I was responsible for the Admin Panel service, which assists in configuring user roles and project permissions. I was also part of the LDLearning team to integrate LiveDesign with the machine learning backend. I revamped the CI/CD pipeline and migrated these services to the microservice architecture using Docker, Jenkins, and Kubernetes for streamlined development, testing, and deployment. Additionally, I implemented ACL enhancements, model retraining, JWT authentication, SSO, and other major features, showcasing my versatile impact on LiveDesign's growth.
Technologies Used: Java, Python, Django, React, JavaScript, Flask, Docker, Kubernetes, RQ.
Enhanced the accuracy of Event Cancellation Prediction, aiming to improve the event ranking process on the company's website. This aimed to bolster customer confidence in the company's service quality while mitigating losses attributed to refund procedures. To achieve this, I analyzed features within historical event data, employing statistical analysis methods, and formulated predictive models utilizing CatBoost and neural networks.
Technologies Used: Python, Jupyter Notebook, OpenCV, Keras, CatBoost.
Developed a web and android application for efficient products search based among millions of products filterable on certain criterias. The application was connect to a nosql database and the whole system was deployed using GCP
Technologies Used: React, Node.js, Flask, JavaScript, Android Studio, Kotlin, MongoDB, GCP.
Code | Website
Designed and engineered a 2D platformer allowing entity posession and their ability usage to navigate complex challenges to complete a level. Conducted comprehensive user testing sessions to collect feedback, and by adhering to Agile methodologies, we enhanced gameplay, resulting in iterative improvements.
Technologies Used: Unity, C#, Google Analytics.
Code | PDF | Video | Play
Executed fine-tuning of the OpenAI Large Language Model (LLM), GPT-3, using a specialized dataset to metamorphose sentences into a polished and considerate formal structure. Introduced a proprietary dataset, with more than 500 instances, attaining fluency and performance on par with human levels.
Code | PDF
Worked on generalizability of 3D medical image segmentation across 10 different tasks and adapting to unseen task without human annotations. Used a U-NET FCN-based model along with 3D image processing techniques, producing results comparable to state-of-the-art.
Code | PDF
Predicting the compatability score of multiple videos to an audio input and sorting the videos according to their appropriateness to the audio. We used the number of youtube views as a metric for our prediction. We extracted various faetures from the audio like scenes, emotions, pacing and matched them to audio features like beats and emotions. We compared and combined these values to get our predicted score.
Code | PDF
For generating intelligent responses with different responses everytime and helping continue the conversation, we built a chatbot using a Sequence to Sequence model. We further improved it using embedding layers for compactness and attention, and teacher forcing mechanism.
Code | PDF