Principal AI Scientist / Architect

Architecting Generative AI that is responsible and production-ready.

I am Joydeb Mondal, a Principal AI Scientist and Architect in Bangalore, focused on Generative AI, Responsible AI, and foundation models. I translate frontier model capabilities into secure, scalable systems, from safety evaluation and model architecture through enterprise deployment.

Get in touch Explore my work
Experience

Leading AI architecture and applied research from strategy to deployment.

2020 - Present Bangalore, India

Principal Applied Scientist

Oracle Cloud Infrastructure

  • Defined Responsible AI platform strategy and shipped guardrails covering PII detection, prompt injection defense, content moderation, and off-topic detection.
  • Built multi-agent safety orchestration with auditable policy enforcement, reducing unsafe prompt success rates by 35-50% and policy violations by 40%.
  • Standardized SLM/LLM training and serving with PEFT, Triton, vLLM, OpenVINO, and hybrid RAG, driving higher throughput and lower inference latency.
  • Deployed Medical BERT clinical NLP models and privacy-preserving PHI detection systems for regulated healthcare workflows.
2017 - 2020

Senior Research Scientist

Nouveau Labs

  • Led the transition from rule-based conversational AI to Transformer-based NLP pipelines, improving intent accuracy by 25-30%.
  • Delivered enterprise NER, contextual response prediction, voice integration, and predictive analytics systems while mentoring junior engineers.
2014 - 2015

Software Developer (ML)

Tecnixs Software & Technologies

  • Developed a machine learning network anomaly detection system achieving 92% F1 on structured traffic features.
Focus Areas

Core expertise for trusted, foundation-model-powered products.

01 / Generative AI

Enterprise GenAI systems

Agentic workflows, retrieval-augmented generation, prompt and context architecture, evaluation pipelines, and production LLM applications.

02 / Responsible AI

Trust and safety architecture

Guardrails, red teaming, model behavior evaluation, policy orchestration, privacy protection, and responsible AI governance.

03 / Foundation Models

Adaptation and scalable inference

PEFT fine-tuning with LoRA and QLoRA, model serving, optimized GPU utilization, distributed training, and domain-aware model deployment.

Research

Selected publications.

ACL 2025

ExpertNeurons at SciVQA-2025: Retrieval Augmented VQA with VLM (RAVQA-VLM)

Mondal, Bhat, Sarkar

NeurIPS 2024

NeurIPS 2024 LLM Privacy - ExpertNeurons

Bhat, Mondal, Sarkar

ACL 2022

ExpertNeurons at FinCausal 2022: Causality Extraction for Financial Documents

Mondal, Bhat, Sarkar, Reza

SemEval 2025

Modgenix at SemEval-2025 Task 1: CAViLR for Multimodal Idiomaticity Understanding

Mondal, Sarkar

Recognition

Edison Award Org-wide recognition for leading ML innovation in OCI AI Services.
Ninja Award Recognition for research output and patent filings across OCI AI.
Nouveau Hero Award Company-wide recognition as performer, contributor, and mentor.
Toolkit

Methods and technologies I build with.

GenAI & Responsible AI

Generative AI Responsible AI Foundation models Safety evaluation Red teaming RAG Agentic systems LLMOps

Frameworks & Orchestration

PyTorch TensorFlow Transformers DeepSpeed LangGraph LangChain MCP

Serving & Retrieval

NVIDIA Triton vLLM OpenVINO FAISS Qdrant OpenSearch Kubernetes

Platforms & Engineering

Python SQL OCI AWS Vertex AI Azure ML FastAPI Docker
Background

Education and continued learning.

Education

MTech, Computer Science IIT (ISM) Dhanbad | 2015 - 2017 | GPA: 8.37 / 10
BTech, Information Technology West Bengal University of Technology | 2010 - 2014 | GPA: 8.14 / 10

Certifications

Building AI Agents and Agentic Workflows IBM Specialization
Agentic AI with LangChain and LangGraph IBM
Advanced RAG with Vector Databases and Retrievers IBM

Leadership

AI architecture and roadmaps Cross-team technical owner for production Responsible AI and GenAI platforms.
Mentorship Mentor to 4+ Applied Scientists and Interns at Oracle Cloud.
Innovation Inventor on multiple AI patents and published researcher.
Contact

Let's architect responsible, production-ready Generative AI.

I work at the intersection of Generative AI, Responsible AI, foundation models, and agentic systems. Reach out to discuss AI architecture, research, or applied AI leadership.

joydeb28@gmail.com +91 7908587846 LinkedIn GitHub