Jun 2025 – Present · 9 mos

May 2025 – Present · 10 mos · San Ramon, California, United States · Hybrid

• 📌 Developed BrightOwl Assistant using Chainlit to create an interactive AI education platform featuring role-specific interfaces for students and educators, implementing real-time chat streaming and JWT-based session management for secure user authentication and chat persistence.
• 📌 Engineered Google Drive integration via custom MCP server using Express.js and Google APIs SDK, building endpoints for file uploads (/gdrive_write_file), document search (/gdrive_search), and automatic chat transcript saving, while optimizing OAuth2 token caching.
• 📌 Deployed Qdrant Cloud vector database with OpenAI's text-embedding-3-large embeddings (1536-dim) and LlamaIndex document processing, implementing hybrid RAG with semantic search and keyword filtering for 10K+ daily queries.
• 📌 Implemented DALL-E 3 image generation with context-aware prompt chaining and automatic Drive persistence, processing base64-encoded PNGs through the MCP pipeline for educational diagram creation.
• 📌 Built Stripe subscription system with SQLite-backed usage tracking, implementing webhook validation and session-based access control for Pro tier features including Claude 3/GPT-4o model switching.
• 📌 Designed specialized AI modes including Deep Research (10K-word academic reports with LaTeX), Math/IIT Math (fine-tuned RAG on curated datasets), and Google Classroom integration via MCP tooling for assignment management.

Apr 2025 – Present · 11 mos

• 📌 Worked on my first course as a LinkedIn Instructor focused on Mixture of Experts (MoE) architecture - covering its mathematical foundations, gating mechanisms, expert routing strategies, architectural variants, training challenges, and how MoE is being actively used in leading Large Language Models (LLMs)
• 📌 Worked on building my second course with LinkedIn Learning on Probabilistic Knowledge Graphs (soon to be published)

Mar 2025 – Present · 1 yr · Cupertino, California, United States · Remote

• 📌 Designed and implemented a graph-based RAG (Graph-RAG) and Hybrid-RAG framework tailored for medical question answering, enhancing the interpretability and trustworthiness of clinical AI outputs.
• 📌 Led efforts to align and integrate biomedical ontologies (including SNOMED CT, UMLS, and MeSH) into a unified Knowledge Graph for downstream QA tasks. Developed ontology-driven entity linking and concept normalization pipelines to map unstructured clinical queries to structured representations.
• 📌 Engineered custom prompt tuning strategies for large language models (LLMs) to improve domain-specific reasoning grounded in graph contexts. Collaborated cross-functionally with medical experts and ML researchers to validate system outputs against gold-standard clinical interpretations.
• 📌 Evaluated the Graph-RAG QA system using LLM-as-a-judge methodologies across seven standard medical QA benchmarks, including MedQA, MedXpertQA, PubMedQA, MMLU-Pro, MedMCQA, MedBullets, and HLE, ensuring robust performance and clinical fidelity.

May 2024 – Sep 2024 · 4 mos · Cupertino, California, United States · Remote

• 📌 Transformed cancer research data accessibility by harnessing the power of Large Language Models to autonomously generate Knowledge Graph triplets from vast repositories of medical literature. By integrating a meticulously defined ontology with semantic reasoning, this approach unveiled intricate relationships, enabling structured, scalable, and intelligent data representation across diverse cancer types.
• 📌 Designed and implemented sophisticated ontologies and Knowledge Graph triplets in Neo4j, revolutionizing the organization, retrieval, and interoperability of medical datasets.
• 📌 Engineered an advanced conversational AI interface for medical Knowledge Graphs by integrating Retrieval-Augmented Generation with Prize-Collecting Steiner Tree optimization. This pioneering framework amplified reasoning capabilities, provided intuitive, context-aware interactions, and enabled researchers to navigate complex medical data with unprecedented precision, significantly advancing decision-making and discovery in the field.

Jan 2024 – Apr 2024 · 3 mos · Ann Arbor, Michigan, United States · On-site

• 📌 Successfully taught STATS 250: Introduction to Statistics and Data Analysis, covering core concepts such as Probability, Hypothesis Testing, Regression, and Analysis of Variance.
• 📌 Led interactive labs and discussion sections, fostering an engaging learning environment while providing personalized support through regular office hours.
• 📌 Designed and graded assessments, organized comprehensive exam review sessions, and ensured fair evaluations with constructive feedback to enhance student learning outcomes.

Aug 2023 – Nov 2023 · 3 mos · Ann Arbor, Michigan, United States · On-site

• 📌 Developed the curriculum for STATS 315: Statistics and Artificial Intelligence in collaboration with Prof. Ambuj Tewari, crafting a rigorous and industry-aligned learning experience.
• 📌 Designed in-depth lecture content and accompanying instructional materials, including slide decks and interactive Jupyter notebooks, covering foundational to advanced topics. The curriculum explored Linear Algebra, Linear Regression, and TensorFlow/Keras, progressing into cutting-edge concepts such as Dense Neural Networks, Convolutional Neural Networks, and Transformers, equipping students with both theoretical depth and hands-on expertise in AI-driven statistical modeling.

Jan 2022 – Dec 2022 · 11 mos · Remote

• 📌 Throughout my role as a Research Assistant, I have been instrumental in pushing the boundaries of semantic intelligence and AI systems across various domains. I developed a methodology to advance knowledge graph curation, leveraging data sources like Wikidata and DBpedia, and employing ontology-driven strategies to set new standards in semantic data representation with remarkable accuracy.
• 📌 In the realm of Social IoT, I tackled challenges in resource discovery and scalability by generating extensive metadata and applying semantic technologies to enhance interactions among smart objects, thereby improving precision and efficiency.
• 📌 On platforms like Quora, I implemented a knowledge-centric approach utilizing machine learning and ontology-based models to accurately detect duplicate questions, thus optimizing information retrieval and enhancing user experiences by minimizing redundancy.
• 📌 Additionally, I engineered an innovative image tagging system for social media, combining machine learning with ontology-driven frameworks to provide precise, context-aware tag suggestions, thereby enriching user interaction and managing content effectively.