Leading and managing AI/ML initiatives, focusing primarily on Generative AI. Driving the development and implementation of innovative AI/ML solutions, leveraging generative AI to deliver cutting-edge products and services. Providing expert consultation and guidance to clients, leveraging my deep understanding of AI/ML technologies and industry best practices. Actively contribute to the AI community through knowledge sharing, presenting at conferences, and participating in industry events. Following are my major contribution since Feb'24.

• Cognitive SDLC: Lead ideator/creator of a custom multi-role multi-agent (MRMA) framework to automate software development life cycle. Features include auto coding, documentation, suggestion, testing, diagram and document understanding & generation and project management operations through plugins (JIRA, Confluence, OneDrive).  Custom and portable solution and can be adaptable to on-premise (local LLMs & local vector store) or provider (Azure/GCP) options. Ideal for rapid prototype development, business intelligence, research and experimentation.  Multiple sucess stories across internal and client use case(s) & fast track prototyping workshops.
• KYC-BOT: KYC solution is for banking and financial services. The chat-based solution is capable of interracting with bancked databases, handling data collision, search and summarize.  Ideal for KYC operators, helps to reduce manual efforts through internal and external verification. The solution is highly customizable and modular and can be adapted to many BFS use cases. 
• Gen AI Powered Intelligent Test Suite:  Co-developed with one of the world-leading banks in Netherlands. This tesing environment is amied for banking environment and specific to data valut related testing. Automatd test case generation with test data, query existing tests and databases, question answering to power designer, test script convertaion and auto documentation. 
• Cognitive Asset Managemnt:  Co-developed with one of the leading financial asset management organization in Netherlands. Master data management (MDM) using Gen AI. Manage, search, deduplicate and streamline multi lingual noisy user generated content in asset managemnt solutions. Azure powered, noise resitant asset managentment with dynamic dashboard generation for reporting KPIs.
• Predictive Maintanance of Manufacturing Pipelines:  Co-developed with one of the world-leading pharma & manufacturing organization in Switzerland & India.  AI powered early warning and exception detection and management system for pharma manufacturing hubs. 
• Master Thesis Supervisons: Supporting our interns from University of Amsterdam, and Vrije University in their master thesis and participating in AI hackathons. 

Lead the development of end-to-end machine learning solutions (Data, Training, Testing and Deployment) and services, employing deep learning models (LLMs with LangChain, Huggingface Models), Databricks and AWS services such as Sagemaker.  Collaborate closely with colleagues to define project scope, establish objectives and deliverables, and manage projects using JIRA, employing both KANBAN and SCRUM methodologies. Demonstrated strong project management skills, overseeing tasks such as data creation, model development, testing, and deployment. Maintained clear and concise documentation throughout the project, facilitating knowledge transfer and ensuring effective collaboration among team members. Consistently met project deadlines and delivered high-quality results, contributing to the overall success of the team and organization.

• Long document parsing using transformer varients: Involved in the development and implementation of automatic document structuring and comprehension techniques, utilizing deep learning models to extract knowledge from lengthy unstructured contexts. Where, I successfully applied Bi-LSTM+CRFs, BERTs, Longformers, and LayoutLMs to extract entities such as titles, authors, sections, references, and drug names from research articles, improving the efficiency and accuracy of our paper Parsing engine.
• Confidence Calibration for Production Models: Developed confidence calibrators for deep learning models, enhancing their reliability and interpretability in real-world applications.
• Citation Structuring & Linking: Led the identification and structuring of references in scientific documents, streamlining the citation process and facilitating easier access to relevant research. 
• Biomedical Entity Detection (NER): Utilized LSTM+CRFs to extract biomedical entities from scientific articles, contributing to advancements in the field of biomedical research and knowledge discovery.

• Collaborated with Accenture Labs and University College Dublin on a joint research project aimed at developing automated Anti-Money Laundering Solutions using NLP and deep learning techniques. Applied NLP, NER, Relation Extraction and Knowledge Graphs to enhance the accuracy of AML investigations, reducing human effort and increasing detection rates.
• Developed a knowledge graph specific to the finance domain, leveraging neural networks and TensorFlow to improve the performance of the automated AML system. Utilized Python and Git for code development and version control, ensuring efficient and organized project management. Presented system demos at conferences, effectively communicating the capabilities and potential impact of the research to industry professionals and stakeholders.
• "NextGen AML: Distributed Deep Learning based Language Technologies to Augment Anti Money Laundering Investigation" @ACL 2018, showcasing the advancements made in the field.

• Developed a decentralized system for NLP and Machine Learning modules, gathering information and intelligence for economic organizations (banks/insurance) to assess the risk and threat associated with clients.
• Applied NLP, NER, Relation Extraction, Aspect-based Sentiment Analysis, and Knowledge Graph techniques to extract relevant information from financial data and identify entities involved in scams, frauds, and money laundering.
• Utilized expertise in the finance domain, including KYC and Panama Papers, to enhance the accuracy and effectiveness of clients Anti Money Laundring System.
• Implemented Neural Nets and Tensorflow for training and deploying machine learning models, achieving high performance and accuracy. Leveraged Neo4j and RabbitMQ for data storage and message queuing, optimizing system performance and scalability.
• Patent: Integrated monitoring and communications system using a knowledge graph-basedexplanatory equipment management approach.

• Engaged in the development of a multilingual search engine on the tourism domain, based on Apache Lucene, as part of the Cross Lingual Information Access System (Sandhan Phase- II) project at Jadavpur University.
• Responsible for generating automatic query-focused summaries and snippets for webpages, developing a query expansion module, and implementing textual entailment-based ranking algorithms. Successfully implemented Java and J2EE technologies, along with Apache Lucene and Nutch, to create a robust and efficient search engine.
• Publications(s):
  • Pakray, P., Barman, U., Bandyopadhyay, S. and Gelbukh, A., 2012. Semantic answer validation using universal networking language. International Journal of Computer Science and Information Technologies, 3(4), pp.4927-4932.
  • Pakray, P., Barman, U., Bandyopadhyay, S. and Gelbukh, A., 2011. A statistics-based semantic textual entailment system. Advances in Artificial Intelligence, pp.267-276.