📄 Know about my experiences https://docs.google.com/document/d/16-b79dI9akCTwfZ06UCMKuj8uOWyMyqD/edit?usp=sharing&ouid=109661486016530485549&rtpof=true&sd=true

• Hi. I'm a 24-year-old Machine Learning Engineer working with NCICS, NOAA NCEI and I have a master's degree in Computer Science(Data Science Track) from NC State University. I've 3+ years of experience designing, implementing, and deploying large-scale generative AI solutions on AWS. Expert in building distributed training pipelines, fine-tuning LLMs, and optimizing ML models. Proven track record of collaborating with cross-functional teams and enterprise customers to deliver customized AI applications at scale.

1. I recently completed a paper with Ensemble for Hallucination Detection. Our study delves into a suite of unsupervised metrics for summary consistency, analyzing their correlations with human evaluations. We've discovered that methods based on Large Language Models (LLMs) like GPT significantly outperform other metrics in detecting hallucinations.

By comparing these metrics to models made from an ensemble, we found that ensemble methods can further improve accuracy, especially when the metrics in the ensemble have similar and uncorrelated error rates. Our proposed ensemble method for LLM-based evaluations shows promising improvements over the previous state-of-the-art, marking a step forward in ensuring the reliability of text summarization technologies. Full Paper

2. I also worked on Image/Video processing which involves Multi-dimensional filtres, Visual perception, Contour and feature extraction, Segmentation, Visual information coding. Open to collabrating if you're working on a similar theme. I've been tweaking around with the associated libraries and tools, to understand the space better HERE

3. All of my projects can be found here at My repositories

4. Ask me anything about Python, R, Statistical Data Analysis, Machine Learning Models, variable analysis and Graph theory

1. Hallucinaton Research : The paper discusses the development of an ensemble approach for detecting hallucinations in abstractive text summarization. Hallucinations, which are inaccuracies or information not present in the source text, pose a challenge for generating accurate abstractive summaries. The paper focuses on unsupervised metrics and examines their correlations and effectiveness in detecting hallucinations. By combining these metrics into an ensemble, the study demonstrates that LLM-based methods are more effective at identifying hallucinations than other unsupervised metrics. I present an improved ensemble for LLM-based evaluations, surpassing the previous state-of-the-art. Source code and applicaiton can be found here

2. Athenaeum: is an application dedicated to connecting you with the books you're searching for. It allows you to search the web and find you the books you seek at a reputable distributor and a good price. Using Athenaeum, you can simplify your journey and minimize your costs as you find the resources you need for class. I have specifically aimed to target students, but all lovers of books are welcome to partake. Source code can be found here

3. Music Genre Classification : With this web application, I aim to classify a clip of audio into one of the predefined genres of music. I have used the K-nearest neighbors algorithm because various research it has shown the best results for this problem. The main idea is to extract features and components from the audio files, which includes identifying the linguistic content and discarding noise. A data set containing 10,000 audio sample files from GTZAN was used to train the K-NN model to classify them into 10 broad categories.
   Source Code can be found here

4. Uber Pickups in NYC : This project aims to analysis the pickup locations of users of uber and make meaningful inferences based on them. Another obkective to simply create representative visualizations for the import features. Source code can be found here

5. Traffic Monitoring System : This is a project designed to extract meaningful data using image processing and use that data to train a HOG/SVM classifier to generate a system to reduce the wait time for cars and pedestrians in high traffic zones. Source code can be found here

6. Find My Roomie : FindMyRoomie is a Web Application that provides a platform for wolves (NC State students) to find roommates of their preference. The stakes are high when it comes to finding your best roommate because this relationship starts with a living relationship 😅. The software is designed using Django, Python, REST API, PostgreSQL, HTML/CSS and Javascript with functionalities that allow you to filter and choose your ideal roommate. But if that is too much work for you, we also provide roommate suggestions based on your preferences! Any NC State student could sign up with their NC State Email address from any corner of the world on our website and begin searching for roommates.. Source code and application can be found here