Parth Pandya

Project: Naboo – AI-Powered Technical Assistance Platform

• Objective: Developed a scalable AI-driven platform to automate and enhance responses to customer queries using advanced ML and cloud technologies.

• Key Responsibilities and Achievements:

• Designed and implemented Naboo, leveraging RAG (Retrieval-Augmented Generation) with AWS Kendra for data indexing and OpenAI GPT for intelligent, context-aware response generation.
• Engineered a decoupled architecture using AWS AppFlow to securely capture and synchronize Salesforce data, ensuring a modular and scalable design.
• Developed backend workflows using AWS Lambda, integrating indexed knowledge repositories from diverse data sources such as S3, PostgreSQL, Confluence, and Google Drive.
• Integrated with Salesforce for seamless case analysis and automated resolution generation, improving case resolution times, and operational efficiency.
• Delivered a robust, flexible platform that empowered support teams, significantly enhancing customer satisfaction through real-time assistance and automation.

• Technologies Used: AWS Services (Kendra, AppFlow, Lambda, S3), OpenAI GPT, PostgreSQL, Confluence, Python, Salesforce Integration, and RAG Architecture.

• Inquiry Management Tool: Developed a recommendation system for routing inquiries to experts using advanced NLP techniques:

• Implemented sentence classification, named entity recognition (NER), and semantic embeddings with models like BERT, GIST, ADA, and Mistral.
• Integrated with AWS OpenSearch (Elasticsearch) to store and query embeddings for efficient expert matching.
• Fine-tuned and pre-trained transformer-based models (e.g., Mistral, GIST, BERT) using SageMaker pipelines with MLOps capabilities for scalable training workflows.

• Automated Data Processing Tool:

• Built a tool leveraging Retrieval-Augmented Generation (RAG) techniques to analyze and extract keywords and mappings from multiple Excel sheets using LangChain, OpenAI, and prompt engineering.
• Designed and implemented an end-to-end embedding generation pipeline using AWS Batch jobs, enabling seamless upload and querying of embeddings in AWS OpenSearch.

• SMARTBIO Product:

• Engineered a system to generate expert biographies using historical inquiry data, leveraging clustering algorithms like K-means and HDBSCAN.
• Applied GenAI techniques with embeddings (e.g., ADA, GIST) to identify keyword similarities, remove irrelevant clusters, and enhance data quality.
• Evaluated the system’s recommendation performance using NDCG@K metrics for precision and relevance.

• Regulatory Domain Project:

• Designed workflows to process unstructured data from PDFs, DOCX, and Excel for regulatory analytics.
• Extracted pricing details from raw formats using libraries like Camelot and AWS Textract.
• Conducted clause analysis and predictive modeling with logistic regression, SVM, and other ML techniques.

Technologies Used: Python, PyTorch, AWS (OpenSearch, Batch, SageMaker, S3, EC2, Textract), Camelot, LangChain, GenAI, MLOps.

• Successfully delivered 2 chatbots to TataSteel as a client for their HR-related inquiries(Internal) and customer-facing inquiries(external)
• Design an intent classifier for the chatbot with more than 85% accuracy in production
• Worked on the OTT platform(EROS) to build their CMS system using Django Framework
• Technology Used: Python, Flask, Machine Learning, Redis, AWS, Django, Docker, Celery

• Developed a product on sentiment analysis, threat detection, and smart categorisation based on the context of the data using Python and ML techniques
• This product was based on traditional ML algorithms like SVM, Naive Bayes, and Logistic Regression and used Python libraries like Spacy, NLTK, Sklearn, etc
• Achieved 94% accuracy in determining sentiment - improving previous accuracy by 15% - and automated threat detection by 95%