Hire Aniruddha B. - Generative AI Architect

* I am currently working as a Generative AI Architect within the Microsoft, contributing to enterprise-scale AI transformation initiatives in the finance domain. * Over the past several months, I have been responsible for designing and delivering AI-driven platforms end-to-end — from architecture to production deployment — while also driving internal AI initiatives and coordinating with vendor teams on implementation and integration. * My work includes: * Designing intelligent incentive platforms powered by LLMs * Building enterprise-grade RAG systems with structured evaluation frameworks * Developing NL2SQL solutions for secure natural language access to financial data * Architecting scalable AI solutions using Azure ecosystem (Azure AI Foundry, Azure OpenAI, Azure Functions, Container Apps) * Implementing CI/CD pipelines, containerization (Docker), and deployment workflows * Establishing guardrails, observability, monitoring, and governance for agentic AI systems * Managing small vendor teams to ensure delivery alignment with architectural standards * In addition to delivery, I actively drive internal AI initiatives focused on: * Responsible AI implementation * Agentic system architecture standards * Monitoring and evaluation frameworks * Production-grade AI deployment patterns * My approach combines technical depth with systems thinking — ensuring AI solutions are: * Reliable * Secure * Observable * Scalable * Business-aligned * I believe enterprise AI is not about experimentation alone — it is about disciplined architecture, governance, and measurable impact.

* AI-Driven Multi-Agent Workflow for Ticket ResolutionDesigned and developed an AI-powered multi-agent system aimed at optimizing JIRA ticket resolution through intelligent automation. The system utilizes multiple agents, each tasked with a specialized agents for ticket analysis, solution summarization, knowledge base retrieval, corrective action evaluation, and even code generationKey Technologies: GPT-4o, Atomic-Agents, Retrieval-Augmented Generation (RAG), Python, Pydantic, FastAPI, ChainlitMy Role:Sole contributor responsible for end-to-end development, integration, and deployment of the solutionLed presentations to both internal teams and clients, demonstrating the effectiveness of AI in reducing ticket resolution time and increasing operational efficiencyAdditional Contributions and Initiatives:RAG Implementation: Contributed to a pilot implementation of Retrieval-Augmented Generation (RAG), including both Agentic RAG and Hybrid RAG models, to enhance the contextual response quality in ticket resolution.Privacy Compliance: Led the creation of a PII detection and evaluation pipeline using LLM (Phi-3) and Microsoft Presidio, ensuring enhanced privacy compliance for ticket handling.Custom JIRA Tool Development: Built a custom JIRA tool for Create, Update, and Delete (CUD) operations, leveraging Pydantic for robust validation and ensuring seamless integration with the atomic-agent architecture.A2A Protocol Pilot: Developed a pilot for Agent-to-Agent (A2A) protocol, enabling autonomous collaboration between agents, which enhanced system efficiency and reduced human intervention.ITOps Automation Agent: Created an ITOps automation agent capable of performing sentiment analysis, issue prioritization, and classification of issue types, streamlining the support workflow for quicker issue resolution.PII Evaluation Metric: Defined and implemented a novel PII evaluation metric for JSON data, improving the reliability and consistency of privacy audits.

* Currently engaged in the creation of a Gen AI Assistant System to help telecom operations in the daily activity for both structured data and unstructured data using large language model to extract best possible answer for a question asked from multisource documents. * Fine Tuned code llama and llama-3 model for text to SQL generation task for some specific use cases along with various prompts. * Harnessing speech-to-text and text-to-text translation models for automatic speech recognition and seamless translation tasks. * I'm actively engaged in prompt engineering work, refining the system's ability to generate responses effectively. * Detect anomalies occurring in Microwave nodes resulting degraded user experience using Autoencoder. * Anticipating and preemptively identifying RRU faults. * Implemented traffic balancing in Mobily's core network, in collaboration with Ericsson Managed Services, boosted operational autonomy during high-demand events. This led to an award-winning solution (AI Ops Award, FutureNet MENA 2024), part of the Ericsson Operations Engine. Using AI, we forecast and dynamically adjusted traffic across core nodes 12 hours in advance, enhancing efficiency, reliability, and user experience. * Implemented custom NER strategy to recognize telecom specific vocabulary with the correct tag using Spacy and BERT based model. * Developed a model using Roberta to identify valid question based on the telecom keywords using NER tag as a feature in the sequence classification model. * Added domain specific keyword to tokenizer vocab to enhance contextual understanding of Roberta model.