PRATISHTHA SONI

Designed and implemented an AI system with RAG integration using AWS S3, Lambda, OpenSearch, and API Gateway, providing agents with customized feedback based on client compliance knowledge articles and datasets, aimed to automate QA processes.

Utilized Prompt Engineering to automate quality assurance and provide real-time feedback to agents by analyzing and evaluating real-time customer-agent call recordings, enhancing response accuracy by 90% and agent performance.

Designed a scalable call analytics platform using AWS S3, DynamoDB, Lambda, API Gateway, and third-party transcription services, with CI/CD pipelines enabling automated deployment to handle up to 5,000 users.

Architected and optimized microservices-based solutions on the AWS cloud, improving agent analytics by increasing query efficiency by 25% and cutting query costs by over 40%. This accelerated real-time feedback delivery and resolved delays in performance reviews.

Technologies used:

• Python
• AWS (EC2, Lambda, DynamoDB)
• Prompt Engineering
• Azure AI Studio, AI services

Developing and implementing Machine Learning Models after collecting images and leveraging pre-trained models and data analysis, using opencv within Tensorflow and Keras frameworks to accurately predict food images for homeless community.

Publication of “Design of Multisensor System for a Smart Cooking Assistant" In IEEE International Conference on Artificial Intelligence x Medicine, Health, and Care (AIMHC) 2024.

Explored diffusion models and multimodal frameworks, leveraging transformers for LLMs, text-to-text and image generation, to enhance image recognition capabilities.

Technologies used:

• Python
• PyTorch
• Tensorflow
• Keras

Teaching Python and C++ to undergraduates with expertise in communication skills and curriculum design.

• Developed and tested APIs for client website integration, optimizing cloud operations, achieving a 50% cost reduction, and improving process debug time for enhanced operational efficiency.
• Engineered and implemented 10+ new features and collaborated effectively with cross-functional teams to enhance user engagement on live-streaming platform.
• Incorporated Agile methodology using Jira and Leveraged cloud-native services such as AWS Lambda, Amazon EC2, and Amazon RDS to build resilient applications, leading to a 30% enhancement.
• Troubleshooted Kafka producer and consumer scripts, deploying them via Git to the production server, thereby resolving issues and ensuring the smooth transmission of live stream data across diverse Kafka topics, effectively managing over 1000 requests.

Technologies used:

• Python
• Kafka
• MongoDB
• Git
• AWS

• Proficient in developing automation python scripts to identify patterns after understanding technical designs of oil & engineering documents like .csv, .pdf, .xlsx and using SQL queries using Microsoft SQL Server by leveraging existing organizational-based knowledge and NLP-based platform to accomplish automation in data extraction from documents to achieve 70% of accuracy and saving 25 hours of manual work per week.
• Developed ELT/ETL framework to transfer data from various data sources, including cleaning, removing outliers, and preprocessing, into both data warehouse and data mart environments for reporting and analytic purposes.
• Independently executed data integration scripts using SSIS to read .json, .xlsx files and automated tabular data extraction by tableau and pandas library and export it to a predefined report format to help reduce manual data entry processes by 40%.
• Developed and maintained a scalable Data Lake Architecture to store 50m+ structured and unstructured records streaming every week.
• Incorporated agile ceremonies to address clients requirements, crafted ontologies and knowledge graphs with python for viewing relationships among 40+ excel sheets in Protégé.
• Utilized tools like Microsoft Power BI, Tableau and SQL queries for conducting statistical data analysis to reveal meaningful patterns in the dataset and for data visualization.
• Proficient in utilizing Python collections to efficiently manage and manipulate large datasets, enhancing data processing capabilities.
• Implemented end to end implementation of the project by incorporating agile ceremonies to address clients requirements, designing, coding, testing ontologies and knowledge graphs with python scripts for viewing relationships among 40+ excel sheets in Protégé

Technologies used:

• Python
• PowerBi
• Flask
• ETL Pipeline
• Docker
• MySQL
• Azure