Abhirup Choudhury

About Abhirup Choudhury

Principal Data Scientist with 7+ years in NLP and applied statistics, expanded since early 2025 into applied AI system prototyping and evaluation for risk, compliance, and public-sector use cases. I operate as a technical catalyst and strategic prototyper: I take ambiguous problem statements and turn them into decision-grade prototypes, evaluation harnesses, and architecture feedback that engineering can scale. Comfortable coding across Python, TypeScript/JavaScript, and Java, I translate complex system designs into demoable implementations that align leadership, product, and sales early—often compressing concept cycles from months to days, and in time-sensitive cases, hours. My work emphasizes rapid validation, failure-mode testing, and guardrails so advanced AI capabilities behave predictably in regulated environments.Core Competencies: • Zero-to-One Prototyping & System Design: POCs, feasibility assessment, demo engineering, design tradeoffs, architecture validation• Applied AI (Agentic + RAG): multi-agent workflows, orchestration patterns, retrieval pipelines, evaluation & test harnesses• Risk & Compliance Analytics: sanctions screening concepts, supply chain resilience, entity & relationship analytics, graph reasoning• Engineering Toolkit: Python, TypeScript/JavaScript, Java, SQL, GCP (Vertex AI), vector databases, graph tooling• Cross-Functional Leadership: requirements translation, stakeholder storytelling, roadmap shaping through experiments, engineering feedback loops• Legacy/Previous Stack: SAS

Experience

1. Principal Data Scientist

   Dun & Bradstreet

   Oct 2025 — Present · Washington, DC, US

   Hands-on Principal Data Scientist supporting an applied AI program for public-sector supply chain risk. Build prototypes and internal tooling, run architecture stress tests, and provide design feedback that reduces delivery risk and accelerates iteration across engineering, product, sales, and senior leadership.Key Responsibilities & Impact:• Zero-to-one prototyping for decision-making: Build working prototypes for complex use cases (e.g, document reconciliation workflows, relationship/contagion risk exploration) to make designs concrete and speed alignment across leadership, product, and customer-facing teams.• Cycle-time compression: Reduce concept-to-POC timelines from months to days, and in time-sensitive cases, hours, by rapidly implementing decision-grade prototypes and iterating through evaluation and stakeholder feedback.• Architecture stress testing & failure-mode analysis: Design edge-case and adversarial test scenarios for agent workflows; deliver actionable feedback to engineering on reliability, performance, and compliance-sensitive failure modes.• Guardrails and policy-constrained behavior: Translate mission and compliance requirements into technical constraints (guardrails, routing logic, middleware patterns) to keep AI outputs consistent with regulated expectations.• Internal tools that increase throughput: Build internal tools that automate research, analysis, and evaluation workflows for peers—reducing manual overhead and increasing iteration speed on prototyping and investigations.• Research-to-roadmap contributions: Identify opportunities, test hypotheses, and share findings via lightweight prototypes, benchmarks, and design notes—informing roadmap direction with evidence and tradeoff analysis.Core areas:Applied AI (RAG/agent workflows), rapid prototyping, Python, TypeScript/JavaScript, Java, SQL, GCP Vertex AI, LangGraph, graph analytics, stakeholder management, evaluation & testing, technical strategy, supply chain risk

Education

• South Dakota State University

  Bachelor’s Degree

  2009 — 2013

• Northern Arizona University

  Master's degree

  2013 — 2015