Dec 2023 – Dec 2025 · 2 yrs · Bengaluru, Karnataka, India · Hybrid

• Watsonx Client Engineering for Americas.
• Leading GenAI Research and Co-creating pilots with the clients.
• ► Led the design and development of an LLM based PL/SQL to Java code translation framework for one of the largest Banks in Canada to strengthen their application modernization efforts - enhancing productivity by 70% and $1.4Mn+ saving in associated labour cost.
• ► Led the design and development of an LLM-based ticketing system for a Canadian Govt. organization consisting of a user chat interface having knowledge base across the data sources, handling multi-bot scenarios, follow-ups, auto-filling of missing information in the tickets created, etc. with >95% accuracy and coverage, saving the client $400k+ in agentic efforts.
• ► Led the process automation efforts for a Canadian Insurance brokerage firm to provide their brokers with improved processes, infused with automation and AI, increasing productivity and compliance in their underwriting process - 65% reduction in time spent in making policy changes and decisions.
• ► Designed and delivered a multi-agent AI system for a fintech startup, enabling natural language-driven financial planning and goal-based insights. Integrated agentic RAG with tools, memory, and reflection - boosting customer satisfaction and retention by 45%.
• ► Engineered and implemented a multi-agent orchestration system for a shipbuilding company to automate and accelerate their ITB client intake, education, and lead qualification process - 23% reduction in workload, 24-hour faster response times, and ∼$1K/client savings in training costs.
• ► Architected and led development of a multi-agent conversational AI solution in TA domain, enabling autonomous candidate sourcing, invitations, and shortlist retrieval - cutting recruiter effort by 60% and delivering over $250K/year in operational savings.
• ► And many more ...

Apr 2022 – Dec 2023 · 1 yr 8 mos

• Built and Managed end-to-end AI products for Zycus's Merlin AI Studio.
• Lead multiple projects in Merlin - an AI platform for Zycus's Procure-to-Pay software.
• Worked on AutoML pipeline to reduce experimentation time. Invoice Predictive Paths to predict e-invoice life cycle. Liveness-detection to detect real/fake subjects in live videos, VM based architecture to Docker-Kubernetes migration and many more...

Apr 2021 – Mar 2022 · 11 mos

• Built and Managed AI products for Zycus's Merlin AI Studio.
• Highlights...
• ► Tax Predictor: Developed an AI BOT that predicts all the combinations of taxing details for an invoice.
• ► Process Mining: Built a system that can mine & predict end-to-end e-invoicing flow including time taken for a step, the reasoning for an invoice to be on hold, etc.
• ► Fraud Detection: Handled an AI module that flags potential duplicate invoices, and detects fraudulent emails.
• ► Multilingual Support: Worked on multilingual support for almost all our AI modules. Engineered a language detection and translation service.
• ► Multi-Tenant Architecture: Devised system for multi-tenancy across modules
• ► Cognitive Analytics: Worked on a Big Data analytical solution for opportunity recommendation across use cases.

Apr 2020 – Mar 2021 · 11 mos

• Built AI products for Zycus's Merlin AI Studio.
• Highlights...
• ► Requisition-Filling: Built an AI Bot that auto-fills requisition requirements & also allows users to make corrections if needed, achieving an F1 score of 96% for 15+ fields.
• ► Meta-data Extraction: Devised an Information Extraction system exploiting 30+ linguistic features for extracting supplier names, company names, addresses, and banking details from invoices, with 93%+ accuracy.
• ► AP Smart Desk: Developed a multi-class text classification bot with multilingual support for emails. Utilized an ensemble of embeddings, topic-modeling, and word-sense disambiguation with BERT modeling to achieve an F1 score of 95%+ for 10 classes with highly imbalanced data.

Jul 2020 – Dec 2020 · 5 mos

• Coaching/Mentoring activities include taking virtual classroom sessions for group of students/professionals on Advance Data Science topics e.g. Classic Machine Learning Algorithms, Deep Neural Networks, Recommender Algorithms, Industry experience (Practical knowledge) sharing, and helping in interview preparations.

Dec 2019 – Mar 2020 · 3 mos · Bengaluru, Karnataka, India

• Training activities include taking weekend classroom sessions for group of students/professionals on Advance Data Science topics e.g. Classic Machine Learning Algorithms, Deep Neural Networks, Recommender Algorithms, Industry experience (Practical knowledge) sharing, and helping in interview preparations.

Dec 2018 – Apr 2020 · 1 yr 4 mos · Bengaluru, Karnataka, India

• Worked on multiple projects in NLP and Computer Vision for financial domain.
• Highlights...
• ► Bounded Box Detection: Designed a component that converts a physical document into a JSON tree which consists of different sections: page, page-header, page-footer, figures, paragraphs, and tables along with its cell structure using various ML/DL tools, improving the OCR quality by 13%.
• ► Relevant Page Detection: Developed a component that detects the relevant pages based on the user label data, used the BERT (Bidirectional Encoder Representations from Transformers) technique for NLP pre-training to get the almost perfect representation of seq-to-seq data, and built a highly generalized model.
• ► Field Identification: Developed a component that extracts values for the fields or data points specified by the clients.
• ► Field Recognition: Developed ML/DL gates for the decision tree, which helps extract field values more robustly and accurately.
• ► Named Entity Recognition (NER): Helped develop a custom Named Entity Recognition model for the financial domain. Used Bi-LSTM, Character Embedding, and Conditional Random Field to develop a highly efficient and accurate model for production.

May 2017 – Jun 2017 · 1 mo

• Internship activity includes scraping publication data from various management journals, collating, cleaning the data, and developing an AI based ranking system.