About me
I recently graduated with my Masters in Computer Science and focus in Artificial Intelligence, encompassing a broad spectrum of relevant courses such as Natural Language Processing, Computational Linguistics, Machine Learning, Distributed Systems, Operating Systems, and Analysis of Algorithms. Exposure to Machine Learning and Computer Vision during my undergraduate years prompted me to delve into the field further. Subsequently, I dedicated 1.5 years post-Bachelors to gaining practical experience in these areas as a Machine Learning Engineer.
Education and Work ExperienceJourney through a diverse spectrum of educational pursuits and professional experiences
Machine Learning Intern, Thrivent[May, 2023 - Aug, 2023]
- Created document summarization and information retrieval services tailored to query medical documents with ChromaDB and LangChain using Dolly 2.0 12b with an average BERTScore of 0.93
- Built model scoring scripts for generating data-drift and concept-drift statistics to monitor mutual fund propensity using DataBricks and orchestrated it with AWS managed Airflow to reduce potential financial losses by preventing erroneous model
- Operationalized the monitoring framework in production using Bamboo, Airflow and Bitbucket, ensuring seamless deployment and integration within the existing system which also notifies stakeholders about mis-classification and performance issues
University at Buffalo, The State University of New York[Aug, 2022 - Dec, 2023]
Masters in Computer Science and Engineering
Relevant coursework - Natural Language Processing, Computational Linguistics, Analysis of Algorithms, Machine Learning, Distributed Systems, Operating Systems
Machine Learning Engineer, Quantiphi[Feb, 2021 - Jun, 2022]
Unified Ride Safety System with IVA (Intelligent Video Analytics) framework
- Engineered BYTEtrack based real time object detection and tracking system to track riders’ actions at adventure park rides to enhance safety with a precision rate of 0.9534 and mitigate the risk of injuries
- Developed Flask APIs to execute real-time inferences using NVIDIA TensorRT and store the annotated images in Redis DB
- Utilized Vertex AI for model tracking and operationalised a 60-70% reusable platform architecture with modular components
Network Auto-Fault Management System
Devised network anomaly detection system using XGBoost algorithm to diagnose 26 specific cyber attacks Explored on-site benchmarked network dataset and experimented with different bagging and boosting algorithms like Random Forest and XGBoost comprehensively and achieved 5.3% improved recall score of 0.976 Mitigated the risk of downtime and data loss by streamlining real time alerts on-premise to responsible stakeholders
University of Mumbai[Jun, 2017 - Jun, 2021]
Bachelors in Computer Science
Relevant coursework - Data Structure and Algorithms, Database Systems, Data Mining and statistics, Applied Mathematics, Advance Database and Distributed Systems, Big Data, Computer Graphics
ProjectsAn art gallery of innovation and creativity
Falcon 40B fine tuned with Q-LORA
Fine Tuned Falcon 40 B with the help of Quanitized LORA using domain-specific psychographic data focused on human psychology to evaluate its adaptability and efficacy within specialized areas of study. Utilized the state-of-the-art RAG method to enhance its functionality further, incorporating LangChain and FAISS as the vector store. Effectively administered computational assets, such as a sole A100 80GB GPU on Runpod, to facilitate these advancements, demonstrating proficiency in maximizing hardware for improving models.
Multipdf RAGbot with MistralAI 7B
Developed a RAGbot capable of querying numerous documents through a file upload functionality integrated into the UI via Streamlit The utilized language model was Mistral AI 7B, operating on a CPU, while employing SentenceTransformers to embed segmented texts from various documents, resulting in an efficient process for retrieving responses The RAGbot achieved an impressive BERTScore of 0.96 despite being a sharded model, operating on an M2 chip with 16 GB RAM. (github)
Github Q&A using Retrieval Augmented Generation (RAG)
Created a LangChain RAG using GPT 4 as the LLM to query github docs which can be used as a chatbot for any set of documents. This can also be integrated with any application to build rich, interactive AI applications that use data as a source (Github)
Programming massively parallel systems
Implemented parallel algorithm in Python and MPI to solve the Knapsack algorithm, leveraging parallel processing with distributed memory for performance gains over sequential code execution (github)
Hallucination Free Dialogue Generation
Designed 2 stage Faithful Dialogue Generator with a hallucination critic using BiLSTM models as Siamese network and T5 model as an Initial Response Generator (IRG) and Faithful Response Editor (FRE) (github)
Pintos Operating Systems
Incorporated priority scheduling, alarm clocks, and interrupt handling on the Pintos skeleton. Enabled argument passing and system calls to execute user programs. Implemented fundamental file system operations with virtual memory (github)
Distributed Systems Algorithms
Implemented distributed Map Reduce Algorithm (both sequential and parallel), Chandy Lamport distributed snapshot algorithm, Raft Consensus Algorithm utilizing GoLang (github)
Research Work Unraveling the intricate realm of Machine Learning and AI
Supervised Research, University at Buffalo
My past Machine Learning experience got me research work in the University and I actively participated in research under the supervision of Dr. Jinjun Xiong. Throughout this engagement, I explored the correlation between the efficacy of object detection and tracking models and the resolution of the microcontroller camera and its processing. I primarily dealt with TinyML models, including YOLObile and FOMO.
Temporal Analysis and Prediction using Machine Learning and remote sensing, University of Mumbai
Using images extracted from Google Earth Engine using ArcMap, water quality parameters such as COD, BOD, and 4 more were isolated by applying various bandwidth filters. The gathered data was then scrutinized across 13 predetermined geographical regions, and by discerning patterns, predictions regarding water quality were formulated. Additionally, recommendations for enhancing water quality were proposed based on the analysis.
Skills My current tech-suite
Programming Languages
Frameworks
Other
Paper Publications
Temporal Analysis and Prediction of Mithi River using Remote Sensing and Machine Learning
Conference - ICCSDF, 2021 (Abstract)
Evaluation and Transformation Analysis of the Mithi River
Conference - IEEE-CCGE, 2021 (Abstract)