 Developed GenAI-Based Solutions: Designed and executed POCs for NavaX GenAI solutions, including a dynamic data extraction model with over 90% accuracy and a case narrative generation system that reduced correction and rewriting time by 60%, both achieving these results with zero training. Engineered NavaX RegIntel Solution: Created and tested NavaX RegIntel, achieving 90% accuracy in data extraction and an 85% reduction in processing time. The solution also enabled knowledge-base interactions with 95% grounding and 90% relevance in responses. Designed and Implemented Prompt Management Solution: Initially developed an in-house prompt management system. Later, to support the full lifecycle of LLMops, explored and devised the implementation strategy for integrating Langfuse. Led R&D Team: Supervised a team of four Senior Data Scientists, guiding research and development efforts across multiple projects.

 Designed and Implemented Document Enrichment Pipeline: Developed a document enrichment pipeline using AWS technology, including AWS Textract for text extraction, Amazon Comprehend for information extraction, and AWS A2I for human review. Achieved 50% accuracy with zero training, with a system designed to improve over time through human feedback. Conducted POC for Advanced Signal Detection: Led a proof of concept (POC) for advanced signal detection, utilizing statistical ROR and deep learning-based anomaly detection algorithms. This innovative solution, for which ArisGlobal has filed a patent, reduced false positives by 42%, increased ‘true’ signal detection by 19%, and improved signal optimization by 39% compared to traditional methods.

1. Researched and Identified data drift detection techniques for text and structured data.2. Researched and identified the solution for bio-Bert model explanations almost in real-time. 3. Researched and removed the representation bias in the bio-Bert and Spacy model.4. Leading a team of 4 Data scientists involved in research & development and improving the performance of existing models.

1. Identified relationship between two named entities by fine-tuning existing bio-Bert pre-trained model. This solution achieved a 97% F1 score on unseen data, which is far better than the previously applied rule-based solution.2. Identified the named entities from Scanned and Digital PDF by training the SPACY model on the medical dataset. 3. Predicted the listedness of disease in a product by using xgboost algorithm. This model achieved 90 % accuracy on the unseen data.

Descriptive Analysis through Tableau.

• Optimized intent classification accuracy by 5% and load time for the existing by 1 second in existing bot.• Identified the rotation angle for an image by using CNN model.• Detected and extracted text from an identify card image by using Open CV, CNN model and PyTesseract library with 85% of accuracy.• Face Detection, Recognition and Verification by using Open CV and CNN model (Using Face Net algorithms).

• Worked upon the stock price trend prediction using the time series data, Market news , global indices performance data in particular segment ,USD-INR rate and technical indicator. NLP (word2vec) is being used for feature extraction from news data and Deep Learning (LSTM) is being used for overall modelling.• Developed the Data Quality Layer which uses the Deep Learning Model and statistical techniques to identify the anomaly records from the given dataset.• Lead Developer in Investment Performance application (Generate mutual fund performance number) which replaces the existing performance system hosted on Sybase DB.This new application removes the Sybase license renewal cost with improved performance.• Reduced the time for merging multiple benchmark data for single fund to 10 times.• Design and developed the model for effective auditing by linking vendor product and our application on common identifier, this has reduced the application support team analysis time to ½.• Developed the framework for loading data to Target Database from different channel (e.g. csv, xml, json, excel, message queue). This framework has reduced the developed time for any data load to ¼.

• Worked on Refactoring of Banking application where the Underlying Database Model has been redesigned (After some denormalization) to improve the performance of the system and it has been delivered in span of 6 months.

• Led the team of 4 members to maintain Manufacturing sector application and improve the efficiency of existing application by redesigning the application architecture (e.g. removing triggers from the model, designing simplified flow in TIBCO).

• Tuned the daily MIS queries to fetch report to 10% of the total time queries used to take earlier. Developed the utilities to load bulk configuration through csv in billing tool used by the client. It reduces the configuration team effort to 1% of total configuration time through screens.