I have passion for building secure, scalable backend systems that integrate AI and machine learning capabilities. I bring together expertise in cloud infrastructure, DevOps automation, and cybersecurity to deliver high-performance applications. From designing intelligent APIs to managing CI/CD pipelines and securing deployments, I thrive at the intersection of backend development, AI integration, and secure DevOps practices.
Currently, I’m a Founding Backend AI Engineer at Adden.AI, where I design and build scalable, ML-powered systems using large language models. This role has deepened my expertise in distributed system design, real-time data processing, and integrating machine learning models into production workflows. I’ve worked extensively on performance optimization, API design, and infrastructure scaling, while collaborating across teams to deliver reliable, AI-driven products that handle complex ad tech data and logic at scale.
At MIDAS Lab, I gained hands-on experience fine-tuning large multimodal models and building end-to-end AI applications using frameworks like LangChain, ChromaDB, and LLaVA. I developed strong skills in dataset design, model evaluation, and prompt engineering for scientific reasoning, while deepening my understanding of error analysis in LLMs, collaborative research, and publishing in top-tier conferences such as IEEE and AAAI.
Outside of work, I enjoy hitting the gym, playing volleyball, jamming on my guitar, gaming, and getting lost in books and movies.
Founding Backend AI Engineer Adden.AI
Design and develop intelligent, scalable backend systems powered by LLMs and machine learning. Built Reporting AI to automate multi-platform ad reporting with real-time dashboards and KPI tracking. Developed Smart Audience AI for predictive audience targeting using deep neural networks and multi-source data analysis. Created Monitoring AI to provide real-time anomaly detection, alerts, and AI-driven root-cause analysis for campaign performance. Engineered Budget AI to reduce advertising cost-per-response by dynamically reallocating spend based on performance insights. Built Custom Audience AI to enhance lead quality and conversion rates by activating enriched contact data for pre-outreach ad delivery.
Undergraduate Researcher MIDAS
Developed an end-to-end voice chatbot which transcribed audio, generated embeddings with LangChain, stored in ChromaDB, processed inquiries and converted back to speech for seamless user interaction.
Introduced and created GeoVQA, a multimodal math dataset, to enhance LLMs capabilities in geometry. Fine-tuned LLaVA models, achieving superior performance over GPT-4 on the PGPS9K benchmark.
We proposed MoRA, a framework that uses GPT-4o to iteratively fix comprehension, conceptual, and calculation errors in LLMs, boosting Llama-3-70B accuracy by up to 16% on benchmarks like SciEval, MMLU, and PhysicsQA.