Research

Sushanta Paul

Researcher in AI for financial crime, customs fraud and trade-based money laundering. Joint Commissioner of Customs at the National Board of Revenue, Bangladesh, with fourteen years across trade, revenue administration and regulatory compliance. Author and speaker.

• ORCID
• Google Scholar
• Web of Science
• ResearchGate
• LinkedIn
• GitHub
• YouTube

Research interests

• Anti-money laundering
• Trade-based money laundering
• Customs fraud
• Customs regulations
• Trade facilitation and compliance
• Fraud detection
• Financial crimes investigations
• Risk management
• Statistical modeling
• Machine learning
• Deep learning
• Graph neural networks
• Natural language processing
• Large language models
• Explainable AI
• Anomaly detection
• Causal inference
• Predictive analytics
• Algorithmic accountability
• AI governance
• Public administration
• Data analytics

Publications

Journal articles

• Macro–meso–micro analytics for trade-based money laundering risk screening in Bangladesh: A conceptual entity-graph extension
  Journal of Economic Criminology, volume 13, article 100244, September 2026 · Sole authorA three-level framework linking mirror-statistics anomalies, entity-level trade networks and transaction screening for trade-based money laundering risk in Bangladesh, with an explicit map of which laundering techniques customs data can and cannot detect.
• Who the Screen Flags: Auditing and Mitigating Structure-Conditioned False-Positive Burden in Anti-Money-Laundering Screening
  IEEE Access, volume 14, pages 129409–129423, 2026 · Sole authorAudits how false-positive burden in anti-money-laundering screening falls unevenly across account structures, and evaluates mitigation at fixed operational alert budgets. Open access, with code and data released under an MIT licence.
• Governing trustworthy AI in critical communications infrastructure: an ethical and regulatory framework for 5G/6G Open RAN and network automation
  Digital Policy, Regulation and Governance, 21 August 2026 · Sole authorEthical and regulatory framework for AI governance in open RAN and network automation across 5G and 6G critical communications infrastructure. Gold open access, CC BY 4.0.

Data and datasets

• Dataset of 400,000 Prozorro public procurement tender records preserved in synchronization-feed order, with derived award-lot tables
  Data in Brief, volume 69, article 113365, October 2026 · Sole author · Open accessData article describing 400,000 Prozorro public procurement tender records preserved in synchronization-feed order, with derived award-lot tables (33,148 eligible award-lots; fixed November–December 2023 set of 32,800), acquisition and verification code, and a pseudonymized, field-minimized snapshot. Dataset: 10.5281/zenodo.23348295. Gold open access, CC BY 4.0.
• Temporal provenance and truncation sensitivity in operational government API acquisitions: data, code, and reproducibility materials
  Zenodo, 15 October 2026 · Sole author

Conference papers

• When the Regulator Cannot Read the Model: Technical Governance Capacity Gaps and AI Safety in Developing Countries
  IEEE International Conference on Responsible Artificial Intelligence (IRAI 2026), Melbourne, 2–5 September 2026 · Authors: Shoumya Chowdhury, Anmita Das, Sushanta Paul · Presentation certificate IRAI26-000306Argues that AI safety in developing countries is mainly a capacity problem rather than a policy-design problem, using Bangladesh’s National AI Policy 2026–2030 as a case. It identifies three structural failure modes (institutional underspecification, compute dependency and regulatory-capture asymmetry) and proposes a five-pillar Minimum Viable Oversight Framework for regulators who cannot yet audit a model but must govern it. Presented on 4 September 2026; proceedings to be submitted to IEEE Xplore.
• Comparing Retrieval-Augmented Generation Configurations under Noisy Retrieval: A Taxonomic Review of Hallucination Behaviour
  IEEE International Conference on Responsible Artificial Intelligence (IRAI 2026), Melbourne, 2–5 September 2026 · Authors: Anmita Das, Shoumya Chowdhury, Sushanta Paul · Presentation certificate IRAI26-000308A PRISMA-style systematic review of 116 primary studies that develops an operator-centred taxonomy of retrieval noise, maps eight representative RAG configurations onto a common pipeline view, and proposes a noise-stratified benchmark protocol. Initial measurements on 200 NQ-Open queries show a scale-dependent inversion between Flan-T5-base and Flan-T5-large. Presented on 3 September 2026; proceedings to be submitted to IEEE Xplore.
• BanglaGovRAG: A Retrieval-Augmented Generation Framework for Legal and Administrative Question Answering in Low-Resource Multilingual Settings
  3rd International Conference on Machine Intelligence and Emerging Technologies (MIET 2026), Noakhali Science and Technology University, 8–9 October 2026; proceedings in Springer, Lecture Notes in Networks and Systems · Authors: Shoumya Chowdhury, Anmita Das, Sushanta Paul · Presentation certificate, paper ID 5274

Accepted

• A Noise-Stratified Benchmark for Auditable Retrieval-Augmented Reasoning on Trade-Based Money Laundering (TBML-NoiseBench)
  Machine Learning (Springer), ACML 2026 Journal Track · Authors: Sushanta Paul, Anmita Das, Shoumya Chowdhury · Benchmark, code and data on Zenodo
• Knowing When Not to Answer: Noise-Conditional Hallucination Risk Estimation and Shift-Robust Selective Generation for Retrieval-Augmented Language Models
  Machine Learning (Springer), ACML 2026 Journal Track · Authors: Anmita Das, Shoumya Chowdhury, Sushanta Paul
• Order-Invariant Retrieval-Augmented Generation: Certifying and Achieving Robustness to Passage Ordering under Retrieval Noise
  24th Australasian Data Science and Machine Learning Conference (AusDM 2026), Sydney, 2–4 December 2026; proceedings in Springer, Communications in Computer and Information Science · Authors: Anmita Das, Shoumya Chowdhury, Sushanta Paul
• Learning from Public Import Refusals: A Temporal Benchmark for AI-Assisted Import Risk Triage under Confidentiality Constraints
  24th Australasian Data Science and Machine Learning Conference (AusDM 2026), Sydney, 2–4 December 2026; proceedings in Springer, Communications in Computer and Information Science · Authors: Sushanta Paul, Anmita Das, Shoumya Chowdhury
• Counting Clinicians Before Trusting Judges: How Rating Dependability Limits the Validation of LLM Judges for AI Counselling
  International Workshop on Artificial Intelligence in Mental Health (AIMH), CARES 2027, NIT Goa, India, 29–30 December 2026 · Sole author

Preprints

• Protecting Customs Revenue with Scarce Audit Capacity: Selection-Robust Targeting of Post-Clearance Undervaluation under Label Bias
  SSRN, 2026 · Sole author

Peer review

Independent reviewer for international journals and conferences in economics, public administration, trade and artificial intelligence, since May 2026. In 2026 I completed nine reviews. Manuscript details remain confidential.

• Applied Economics (Taylor & Francis), 1 manuscript, October 2026, with a 2026 Reviewer Certificate
• Journal of Shipping and Trade (Springer Nature), 1 manuscript, October 2026
• International Journal of Public Administration (Taylor & Francis), 1 manuscript, August 2026, with a 2026 Reviewer Certificate
• Digital Policy, Regulation and Governance (Emerald), 1 manuscript, May 2026
• AusDM 2026, 24th Australasian Data Science and Machine Learning Conference, Late-Breaking Work track, 2 submissions, September 2026
• CICAI 2026, CAAI International Conference on Artificial Intelligence, 3 submissions, June 2026

Review records: Web of Science ResearcherID QIR-5518-2026 and ORCID 0009-0007-8071-6971.

Experience and education

Distinction

First in Merit, 30th BCS Examination, Bangladesh Public Service Commission (November 2011). Ranked first among all candidates nationally in the 30th Bangladesh Civil Service Examination.

Certificates

Skills

• Data Analytics
• Machine Learning
• Anti-Money Laundering
• Customs Regulations
• Public Administration
• Peer Reviews
• Research Skills
• Scientific Writing
• Academic Writing
• Creative Writing
• Public Speaking
• Python
• Data Analysis
• Data Science
• Artificial Intelligence
• Bengali
• Regulatory Reform
• Trade Policy
• Compliance Training
• Public Sector
• Financial Risk Management
• Motivational Speaking

Writing and speaking

I also write in Bengali on literature, psychology and human life, and speak to students on careers and study. My essays, poetry and talks are on this site, and the Career Adda talk series is on YouTube. To invite me as a speaker, see Invite Me As Speaker.

Contact

For research collaboration, reviewing or speaking invitations, write to hello@sushantapaul.com or use the contact page.