Hire Dip D. - Generative AI Architect

Ascendion is a software engineering firm delivering enterprise Generative & Agentic AI solutions. Led architecture and production deployment of two large-scale AI systems across Telecom and Healthcare enterprises. ● Led cross-functional engineering teams to Architect and Deploy a unified Agentic AI chatbot microservice on AWS EKS within Charter Communications’ Tech Mobile application, serving 80,000+ field technicians, supervisors, maintenance engineers. ● Architected LangGraph-based orchestration to route role and intent aware queries across task-specific agents (Job Insight, Tech Assist, APIs) with Langfuse observability; reduced support calls by 60%, accelerated service completion with million-dollar savings. ● Design and led an Agentic AI solution (LangGraph) for CareSource client to automate Discharge logs fax intake using Azure Document Intelligence and LLM validation with Human in the Loop, integrating eligibility and Prior Authorization APIs to reduce manual review; improve auditability. ● Led serverless deployment with Azure Functions & CI/CD; implemented observability via Application Insights and centralized prompt versioning, logging. Oversaw React-based dashboard deployed on Azure Web App to enable business review & HITL validation. ● Owned end-to-end solution architecture, client demos, serverless and AWS EKS deployment in Dev/Prod clusters, CI/CD pipeline for microservices and infrastructure management; perform code reviews, guide technical design, and lead hands-on POCs. ● Collaborate closely with clients on delivery management, roadmap planning, KPI tracking, business use case definition; advise stakeholders on build/buy/rent strategies, LLM and embedding model selection, vector database & observability tools. ● Led development of a RAG service by extracting and embedding 1,000+ confluence pages and 5,000+ technical attachments (images, pdfs, excel, word) into PGVector DB, preserving document hierarchy and multi-level contextual relationships.

Part of CGI US IP AI team– PulseAI (Hyper Automation, Decision Intelligence, Conversational AI) https://www.cgi.com/en/solution/cgi-pulseai. Pulse Gen AI Studio features: (a) Ask Data (chat assistant service - RAG with various data connectors, code compose, document Q&A (b) Ask Insight (natural language to sql). Self-hosted open source LLM models (Llama2 70B, Vicuna 13B) with Model as a Service architecture deployed in Azure cloud with containerization (docker, kubernetes pods). Team has completed POC’s of Pulse Gen AI Studio with various internal/external clients. ● Currently working with Fannie Mae, developing the road map for Enterprise Generative AI architecture and LLM implementation at various business units & build accelerators and productize their use cases – multi-family chatbot, knowledge management assistant ● Design strategy & technical deployment architectures for (a) Vectorized Data layer - Advanced RAG, PEFT, LoRa; evaluate/recommend Embedding models (Cohere, Titan) & Vector DB (OpenSearch, Milvus) (b) Foundational LLM model garden layer: Llama3, Mistral, Claude, AWS Bedrock/Jumpstart, LLM Agents (c) Control & Interception layer: Guardrails -Protect AI, Governance/Auditing, RLHF, Prompt response analytics, Monitoring/Logging & Cost management ● Develop accelerators, whitepaper, reference architecture for use cases –Text Summarization on Fannie Mae docs (methods - Map reduce, Refine) using Llama-3, Mistral LLM with Langchain & Fine Tune (PEFT -Domain adaption/Instruction) Llama-3 in Jumpstart ● Work with vendors: OpenAI to enable ChatGPT Enterprise/API platform in Fannie Mae; build solution architecture/patterns, work with security, compliance/risk & legal team internally. Evaluate onboarded LLMs with benchmarks -MMLU, HellaSwag etc. ● Set up project structure, environment/team guidelines– change management to improve current development/delivery process to stakeholders, reduce technical debts; influence business strategy and achieve AI/ML maturity

One Concern is a Data & Insights Risk Analytics Tech startup which brings disaster science with Machine Learning for better decision making. It quantifies Resilience from catastrophic perils, extreme weather & climate change into a Resiliency score and empowers business to measure, mitigate & transfer risk. As part of Research & Solutions team (reporting to CSO), we work with external clients (Insurance, Banking, Financial services, CRE, Asset management), Engineering, Go to Market & Sales team to build POC’s for clients and productize/scale data science solutions ● Unique market offering: Resilience score & Business Interruption score of all properties in US & Japan– delivered with API. ● Evaluate hazard, exposure ML & Graph Network models for various perils: flood, hurricane, seismic on a property & impact of surrounding lifeline networks & supply chain (highways, airports, ports, bridges, substations, buildings & residential homes) ● Proprietary vulnerability models predict downtime, recovery time, damage ratio in disaster events with 2035/2050 climate scenarios (RCP 4.5/8.5, SSP 245/585). ● DNA (Data & Analytics) product: Develop Exceedance probability & Downtime statistics pipeline (production code) across multiple hazards, planning horizons, return periods, build wireframe & scale Resilience metrics with Engg team for 20M properties in US/Japan ● Mentor & train fellow data scientists, new hires/interns; do code review (Github). Support pre-sales/post-sales efforts for clients (used Tableau). ● Develop Resilience Adjusted Valuation & Financial Loss model pipeline (production code) using Discounted Cash Flow & Monte Carlo simulation, build wireframe & architecture. Work with product/architects. ● Develop model quantifying supply chain risk for Business Interruption using Graph networks and power simulation network model. Software : Python, Snowflake, Streamlit, GCP, Postgres SQL, Argo workflow