Built a multi-modal evaluation system for LLaMA Visual2Code models within Meta Superintelligence Labs (MSL) using Python (asyncio, multiprocessing), scoring generated code from chart images via weighted semantic + visual similarity metrics across 4,800+ evaluations and 22 chart types, improving benchmark reliability

● Designed and implemented reward calibration logic using DINO-based similarity signals in Python, integrating with RL training loops to increase sensitivity of feedback signals and improve model quality across iterations

● Optimized large-scale dataset querying (500M+ rows) using Presto + partitioning strategies, eliminating out-of-memory failures and unblocking GPU-based training/evaluation pipelines

● Developed high-performance data ingestion systems in Python leveraging async I/O, atomic file writes, and parallelized subprocess execution (curl-based), reducing I/O overhead by 80%+ for the 1.5M+ row MMLottie dataset

(Contracted via TEKsystems)

Working in the Consumer & Community Bank division to deliver scalable backend solutions and build resilient microservices supporting high-volume financial applications to detect and prevent fraudulent account activity.

• Developed and maintained RESTful APIs using Java, Spring Boot, and Core Java for real-time fraud detection services, which improved transaction monitoring accuracy and processing speed.

• Deployed containerized microservices using Jenkins pipelines and AWS tools like EC2, Lambda, S3, and CloudWatch, ensuring high system availability and reduced deployment errors.

• Optimized SQL queries and indexing strategies across PostgreSQL and Oracle databases, which helped lower response latency for fraud risk scoring models.

• Collaborated with DevOps, fraud analytics, and QA teams in Agile sprints to deliver secure, compliant features and conducted system testing to enhance user trust and system stability.

• Improved internal Search Engine Tool with crucial code changes in the Java Spring Boot framework while working in production environments and leading successful deployments

• Drove performance testing on JMeter to see how efficiently the search engine service ran and analyzed key metrics on Datadog to understand impact

• Proactively enhanced the Asset Management Tool UI with Angular, HTML, CSS, and

JavaScript programming, further increasing partner use

• Led initiative among the team in writing and debugging jUnit and Cypress tests for component and end-to-end testing to better client experience and usability

Founded a consulting company which provides Social Media Marketing & Branding and Cryptocurrency Investing Services to clients.

Founded and app developed for SB7, a tech startup which develops, designs, and creates Amazon Alexa Skills and Android apps for the Amazon App Store. I work on IOS apps, web and additional mobile apps. I created merchandise for an e-shop, as well. I oversee the operations of all the production/development and help with designs. I also collaborate with other small companies and utilize social media to promote our products.