PRABIR KUMAR DUTTA

Spearheading the architectural design and deployment of Generative AI Agentic Workflows tailored for high-stakes environments in Banking, Finance, and Taxation. Bridging the gap between cutting-edge LLM development and Responsible AI to deliver secure, ethical, and scalable enterprise solutions.

Key contributions:

• Agentic Ecosystem Development: Architecting multi-agent systems designed to automate complex financial reasoning and tax compliance, utilizing RAG and GraphRAG to ensure data grounding and "zero-hallucination" accuracy.
• Responsible AI Leadership: Established and lead the AI Governance framework, implementing rigorous protocols for bias mitigation, transparency, and ethical oversight in alignment with global financial regulations.
• Rapid Prototyping & Scaling: Leveraging the AIRIA.ai low-code/no-code platform to accelerate the development lifecycle, enabling the swift transition of AI concepts into production-ready financial tools.
• Offensive Security & Red Teaming: Directing vulnerability assessments and red-teaming exercises for LLMs to defend against prompt injection and ensure robust data privacy.
• Domain-Specific Innovation: Translating intricate banking and taxation requirements into technical AI roadmaps, ensuring that agentic outputs meet the stringent auditability standards of the BFSI sector.
• Cross-Functional Orchestration: Acting as a bridge between technical engineering teams and executive stakeholders to align AI initiatives with long-term business sustainability and futuristic technology goals.

Career Progression: System Engineer → IT Analyst (ITA) → Assistant Consultant (AST) → Associate Consultant (ASOC)

Contribution:

AI Solution Architect: Enterprise Atlassian Intelligence

Situation: US based retail brand faced data fragmentation across Jira and Confluence, hindering decision-making for a large data technical ecosystem.

Task: I architected a stateful AI assistant utilizing a self-hosted LLM and Vector DB design to automate reasoning over enterprise data silos.

Action: I deployed Mistral Large 3 via vLLM and Milvus within a private VPC, implementing prefix caching and reasoning items to deliver context-aware "opinions" on project blockers.

Result: Successfully scaled the solution to production, achieving a 40% reduction in ticket resolution time and significant operational cost savings.

AI/ML Solution for Global Telecom B2B Platform

Situation: A global B2B telecom platform required advanced analytics to mitigate churn and optimize customer usage patterns.

Task: I led the development of predictive ML models and structured LLM workflows to automate retention strategies.

Action: I built Python models for behavior prediction, engineered Chain-of-Thought prompts, and containerized the stack via Docker on Azure ML.

Result: Successfully deployed a scalable, multi-environment solution that improved prediction accuracy and streamlined data pipelines.

Retail Analytics B2B Platform

Situation: A B2B retail platform faced significant margin erosion and frequent stockouts due to manual inventory adjustments and static pricing across a 1TB product catalog.

Task: I led the evolution from legacy analytics to an autonomous Agentic AI framework to optimize demand forecasting and supply chain execution with minimal human intervention.

Action: I built Python-based forecasting models and architected a multi-agent system using LangGraph for stateful, cyclic orchestration. I designed "Planner" and "Executor" agents leveraging Chain-of-Thought reasoning and standardized function calling to trigger autonomous stock reorders and dynamic pricing updates. I integrated a RAG-powered BI agent using vLLM for high-throughput, context-aware Q&A, ensuring data sovereignty via a private VPC deployment.

Result: Successfully deployed the framework, resulting in 30% faster inventory turnover, a 20% reduction in stockouts, and a 12% increase in gross margins across all pilot categories.

Other iOS & Cross-Platform Projects

• Led design and development of a native iOS application for a major Indian aviation brand.
• Established MVVM and VIPER architectural frameworks to improve maintainability and debugging efficiency.
• Managed a cross-platform mobile app project for an Australian mining firm using Xamarin.Forms and .NET APIs.
• Implemented AutoFac-based dependency injection within the MVVM architecture.
• Oversaw a mobile shopping app for a UK retail chain, incorporating QR-based payments with offline capabilities.
• Supervised a diverse developer team as Tech Lead, applying MVC architecture with Swift.
• Directed senior engineers in building a native iOS app for a Canadian insurance brand with a strong focus on UI quality elements and business logic.
• Supervised creation of native iPad app for an Asia Pacific Tobacco Company, incorporating offline capabilities and mobile printing solutions.
• Managed project timelines and resources, ensuring timely delivery within budget constraints.

Contribution:

• Situation: Irish local councils needed a unified mobile channel to disseminate timely public notifications.
  Task: Design a scalable, location-aware alerting application.
  Action: Architected and developed an iOS app using Objective-C and SOAP APIs, implementing geotag-based preference management and multi-channel notifications (SMS, email, social, in-app).
  Result: Enabled targeted alerts for thousands of users, improving notification relevance and citizen engagement.
• Situation: Stakeholders required mobile-first data visualization for investment analytics and healthcare monitoring.
  Task: Build intuitive iPad applications for complex data tracking.
  Action: Developed two iPad apps with interactive graph dashboards to visualize investment statistics and monitor nursing-home resident health metrics.
  Result: Improved data readability and decision-making efficiency by ~30% for operational teams.
• Situation: Multiple consumer applications required offline support, geolocation, and real-time engagement.
  Task: Deliver high-quality iOS apps under tight Agile timelines.
  Action: Built several Objective-C apps including offline phonebook (SQLite), location-based guidance apps, event management with REST APIs and APNS, and animated children’s apps using SpriteKit, following Agile practices.
  Result: Delivered 100% of applications on schedule, with reliable offline access and enhanced user engagement.

Contribution:

Situation: Existing crossword game needed expansion to macOS and richer interaction on iOS to boost engagement.
Task: Extend the app to desktop while enhancing mobile user experience and performance.
Action: Developed a macOS desktop version with platform-specific features, enhanced the iPhone app with gesture-driven interactions (drag-and-drop) and animations, integrated Game Center, and optimized code through performance analysis.
Result: Increased user engagement, improved gameplay responsiveness, and expanded the product’s reach across Apple platforms.