Projects

AnimalCare

• Crafted zero-shot object detection pipeline with Meta's Sam2, utilizing Jupyter Notebook for segmentation to isolate animals in video frames.
• Designed an interface with React, Flask, HTML, and CSS for video uploads and processing.
• Optimized AI chat window with Python and REST API, analyzing animal videos only on frame changes for cost efficiency and accuracy.

HealthTech

• Engineered a HealthTech platform with a login page, featuring an Event Agent [ReAct Search] to find nearby events.
• Developed a Training Agent [CoT] using a multimodal LLM to provide personalized training plans from images and text.
• Designed a Nutrition Agent [RAG] that searches PDFs to recommend ready-made health plans.

HouseFinder AI

• Developed specialized real estate platform with a team of 4 to categorize properties with RentcastAPI based on personal preferences like safety and walkability.
• Deployed custom GPT bot to answer more questions the user has.
• Increased app accessibility by implementing WhatsApp API and having a website frontend so the client can send messages through two platforms.

HousePrice AI

• Predicted real-time bay area house prices based on factors such as sq ft,
• Implemented multivariable polynomial regression.
• Utilized machine learning principles such as regression, normalization, and feature engineering for a more precise output.

Medical Scribe

• Implemented real-time transcription with Deepgram to capture patient interactions accurately.
• Employed AWS Medical NLP to extract ICD-10 codes from transcriptions, converting data into structured information.
• Utilized MultiOn to auto-fill medical forms with patient details and diagnostic codes, streamlining documentation.

Detect Seam

• Architected Java image processing application with FIGMA.
• Achieved O(1) access time to enhance project speed.
• Utilized Java's AWT and I/O operations to improve user experience.

Smart Flashcards

• Utilizes Machine Learning for sentiment analysis (e.g., Simple and ML classifiers).
• Dynamic state transitions based on user input with affirmation checking.
• Leverages features like interactive console and file reading for card decks.

Phishkill

• Developed a comprehensive application using Flask, Heroku, and APIs to analyze emails and calculate phishing risk percentages.
• Launched a live, user-friendly application accessible online with a streamlined landing page for easy use.
• Incorporated effective techniques/algorithms to evaluate email content, providing a reliable phishing probability score.