خبراء الذكاء الاصطناعي

Pronob Kumar Barman

Applied AI Scientist

USA

Apprenticeship On-site

نبذة عني

Applied AI Scientist with senior-level experience translating research into impact across healthcare, finance, and cybersecurity. Led development of a PPO-based recommendation engine that increased session retention 15% …

الخبرة

Applied AI Scientist – Behavioral Health

Technuf LLC · Rockville, MD

Dec 2025 - Apr 2026 · 4 أشهر

Designed a PPO-based reinforcement learning recommendation engine that increased session retention by 15% and user engagement by 10%, directly improving patient adherence across 5 behavioral health programs. Formulated multi-metric evaluation protocols benchmarking 4 model variants across engagement, fairness, and retention, reducing model selection cycle time by 30% and establishing a reusable experimentation framework. Operationalized responsible AI auditing using SHAP-based explainability and Fairlearn demographic parity analysis, reducing bias gaps by 22% across 3 protected demographic groups. Led cross-functional collaboration with 3 clinical teams and data engineering to deliver real-time personalized inference at sub-350ms latency, supporting 10K+ concurrent user sessions. Directed compliant data pipeline design processing over 1M behavioral health records daily with 99.7% data integrity, ensuring research reproducibility under HIPAA and GDPR constraints.

AI Research Scientist, Edge Intelligence

Axiado Corporation · San Jose, CA

Jul 2025 - Nov 2025 · 4 أشهر

Pioneered lightweight transformer-based temporal modeling for embedded cybersecurity, improving early-stage threat detection accuracy by 20% and reducing false positive rates by 12% on ARM Ethos-U NPUs. Conducted systematic model compression research using INT8 quantization-aware training and structured pruning, achieving 4x size reduction while maintaining less than 1% accuracy degradation. Reduced edge-device inference latency by 15% through algorithm-hardware co-optimization on ARM Ethos-U65 and Ethos-U85 under 256KB memory and 50mW power constraints. Validated 99.9% parity between cloud-trained and edge-deployed models across 3 threat-detection benchmarks through rigorous cross-environment evaluation protocols.

Applied AI Scientist – Generative AI and Threat Intelligence

Technuf LLC · Rockville, MD

Jun 2024 - Aug 2024 · 2 أشهر

Designed retrieval-augmented generation (RAG) pipelines integrating 50K+ threat intelligence documents, accelerating security incident response by 40% and reducing mean-time-to-resolution by 25%. Developed hybrid threat-scoring models combining probabilistic ranking with domain heuristics, reducing low-value alerts by 30% and improving analyst decision throughput by 35%.

Graduate Research Scientist and Instructor – Healthcare AI

University of Maryland, Baltimore County · Baltimore, MD

Jan 2022 - May 2024 · 2 سنوات 3 أشهر

Invented the PL-NCF (Pseudo-Label Neural Collaborative Filtering) framework to address extreme interaction sparsity in online health community recommendations, achieving a 2x improvement in HR@5 (from 2.65% to 5.30%) and 83.4% relative gain in NeuMF-PL variant. Discovered and formally quantified the separability-accuracy paradox, identifying a significant negative Spearman correlation (rho = -0.59, p

Director, AI Strategy and Analytics

Bangladesh Bank (Central Bank) · Dhaka, Bangladesh

Apr 2018 - Dec 2021 · 3 سنوات 7 أشهر

Led national AI modernization initiatives deployed across 12 regional branches serving 170M+ citizens, integrating machine learning into financial auditing and monetary policy analytics, reducing manual audit hours by 5%. Designed anti-money laundering (AML) anomaly detection systems using XGBoost with rule-based flagging, processing $1M+ in daily transactions and identifying over 200 high-risk cases with a 92% precision rate. Developed a Bengali-language NLP conversational AI system handling 3K+ monthly queries, reducing central bank support ticket volume by 25% and average resolution time by 40%. Built time-series forecasting models improving inflation projection accuracy by 18%, directly informing quarterly monetary policy decisions reviewed by senior bank leadership. Provided strategic AI governance oversight for 8 compliance-sensitive financial systems, establishing model risk assessment protocols adopted as institutional standards across 4 departments.

المهارات

Neural Collaborative Filtering Probabilistic Topic Models gDMR gSTM Generative AI Retrieval-Augmented Generation RAG Reinforcement Learning PPO Recommendation Systems Anomaly Detection Transformer Architectures Fairness-Aware Machine Learning Demographic Parity Auditing Explainable AI SHAP LIME Causal Inference Model Governance HIPAA Compliance GDPR Compliance Bayesian Inference Hypothesis Testing Spearman Correlation Analysis Cohen Kappa Thematic Analysis Quantization-Aware Training Structured Pruning Evaluation Protocol Design HR@K NDCG Python PyTorch TensorFlow scikit-learn SQL Spark C/C++ AWS SageMaker MLflow Docker FastAPI ARM Cortex Ethos-U NPU Optimization XGBoost Node2Vec LDA DMR STM Fairlearn