About Me
Machine Learning Engineer with expertise in object detection, natural language processing, and cloud deployment. Passionate about leveraging AI to solve complex problems and improve efficiency. My interests span across various domains of AI, including:
- Computer Vision
- Natural Language Processing (NLP)
- Large Language Models (LLMs)
- AI Agents
- Deep Learning
- Cloud-based AI Solutions
Experience
Machine Learning Engineer - Parspec.io
Oct 2024-Present | Bengaluru, Karnataka
- Cooking Models
Machine Learning Engineer - Pibit.ai (YC 21)
Jun 2023-Oct 2024 | Bengaluru, Karnataka
- Engineered the Loss Run product through strategic label finalization, annotation planning, and Object Detection model training (YOLO+OCR), improving processing speed by 40% and reducing manual review time by 50%.
- Automated data preparation and model training pipelines on Azure Machine Learning, reducing data preparation time from 3-4 hours to 30-45 minutes. Implemented versioning and logging for data assets.
- Optimized YOLO model performance by 35% through extensive experimentation with dataset variations, model architectures, and hyperparameter tuning, enhancing object detection accuracy in production systems.
- Deployed models as a service using on AWS CodePipeline on AWS EKS and Lambda with 300ms latency.
- Developed 3 identity services by integrating YOLO and GPT-3.5, achieving 0.85 mAP and 0.95 accuracy at the document level using small sized object detection models.
- Orchestrated the experimentation and testing of GPT-4V capabilities for Document Extraction; directed 4 interns in rapid experimentation, enhancing duration 25%.
- Boosted existing system efficiency through strategic implementation of advanced prompting techniques, including zero-shot, few-shot, chain-of-thought, and sequential prompting methodologies.
Deep Learning Intern - Pibit.ai (YC 21)
August 2022-Jun 2023 | Bengaluru, Karnataka
- Enhanced an existing object detection model (YOLOv5) by implementing image weight sampling, improving the mAP score from 0.62 to 0.74.
- Integrated Azure MLFlow for better experiment as well as model tracking and built an auto-annotation pipeline using CVAT data annotation tool.
- Trained an OCR-free document model (DONUT) with an accuracy of TED 0.81.
- Implemented multiple YOLO models for identity services on AWS Lambda with mAP score of 0.9.
- Worked on template standardization to handle imbalances and data drift for multiple document templates.
Projects
DogLLaMA
- Fine-tuned the LLaMA2-7b language model on an English text dataset translated into Dog Style Messages using GPT-3.5-turbo, creating a unique and playful model called "DogLLaMA-7b" in llama2 compatible format.
- Published the DogLLAMA dataset on Hugging Face.
- For model fine-tuning, utilized PEFT with LoRa and BitsAndBytes for quantization.
BudgetGPT
- Engineering Budget QA Agent using GPT4o-mini and RAG technologies for the Indian budget-related inquiries.
- Optimized data retrieval by implementing an in-memory vectorstore for OpenAI Embeddings of each PDF chunk.
- Utilized the concept of tools to use different vector indexes based on the user query.
- Orchestrated an Agentic RAG system using LlamaIndex and deployed on streamlit.
Blogs
Coming soon...
Stay tuned :)
Skills
Frameworks
Tools
Infrastructure
Cloud Services
Databases
Education
B.Tech in Computer Engineering
K.J.Somaiya College Of Engineering, Mumbai, Maharashtra
August 2019-May 2023 | CGPA: 9.35
Coursework: Data Mining and Analysis, Software Engineering, Operating Systems, Artificial Intelligence
Contact Me
If you'd like to get in touch, feel free to reach out via email or connect with me on LinkedIn.
Email: pathikghugare13@gmail.com
Linkedin: Pathik Ghugare
X or Twitter: pathikghugare