Secure & private AI
Privacy-aware learning, federated systems, and practical safeguards for deploying AI over sensitive enterprise data.
- Privacy-aware learning
- Federated learning
- Model security
Machine Learning Engineer · Deep Learning · Reliable AI
Ajeet Kumar Singh
I build machine-learning systems that perform reliably in the real world.
I work across end-to-end ML/DL—from model development and evaluation to deployment—with deeper expertise in computer vision, robustness, privacy, and responsible AI.
- Focus
- Applied ML & deep learning
- Affiliation
- Research Engineer, TCS Research
- Based in
- Pune, India
Research agenda
Making intelligent systems worthy of trust.
My work connects security, learning, and reasoning—from protecting distributed models to building privacy-aware systems for foundation models.
Robust machine learning
Methods that help models withstand adversarial inputs, unreliable participants, and the distribution shifts encountered outside controlled benchmarks.
- Adversarial robustness
- Byzantine resilience
- Reliable classification
Language, vision & reasoning
Systems that connect text, images, structured knowledge, and databases to answer questions and support more capable human–AI interaction.
- Large language models
- Visual question answering
- OCR & document intelligence
Selected work
ML systems designed for real-world constraints.
Three strands of work show how I move from modeling to evaluation—and where reliability, privacy, or security becomes essential for deployment.
Vision & document AI
01
Teaching visual systems to read before they reason.
OCR-VQA and text-KVQA connect scene-text recognition with visual question answering and knowledge reasoning. OCR-VQA is the most-cited publication on my Google Scholar profile.
Reliable distributed ML
02
Making federated learning resilient to unreliable participants.
FLOT uses optimal-transport barycentric aggregation to improve federated learning under malicious clients and heterogeneous, non-IID data—the conditions deployments actually face.
Responsible LLM systems
03
Evaluating the risks that appear when LLMs leave the lab.
My recent work examines machine-generated text detection and the practical challenges of adopting LLMs across organizations of different sizes.
News
Recent research milestones.
A concise record of recent work, with the complete history preserved on Google Scholar.
-
Optimal Transport Barycentric Aggregation is published in IEEE Transactions on Big Data.
-
No Size Fits All is published at the COLING Industry Track, studying how the challenges of deploying LLMs change with organizational scale.
-
LLMs with Industrial Lens surveys the challenges and opportunities involved in adopting large language models across industrial settings.
-
The TrustAI team reports multi-domain machine-generated text detection results at SemEval, co-located with NAACL.
Publications
Selected papers from Google Scholar.
A selection of recent and foundational work drawn from my publication record. For the complete, current bibliography, visit Google Scholar.
OCR-VQA: Visual Question Answering by Reading Text in Images
International Conference on Document Analysis and Recognition (ICDAR)
A visual question-answering approach and dataset for answering questions by reading the text embedded in images—the most-cited work on my Google Scholar profile.
Patents & professional impact
Research translated into applied systems.
My work spans peer-reviewed research and intellectual property across document intelligence, privacy-preserving ML, image retrieval, and biometric security.
- 1,200+
- Google Scholar citations
- 11
- h-index
- 5
- Patents & applications
Google Scholar profile metrics, August 2026.
-
Patent application · 2022US App. 17/660,034
Method and system for feature-based image retrieval
-
Patent application · 2022US App. 17/361,375
Enabling privacy in an application using fully homomorphic encryption
-
Issued patent · 2020US 10,691,884
System and method for cheque-image data masking
-
Patent application · 2020US App. 16/572,217
Method and system for biometric-template protection
-
Issued patent · 2019US 10,496,894
System and method for text localization in images
About
Engineer by practice. Researcher by training.
I am a Research Engineer in the Cybersecurity Lab at TCS Research, Pune. I work across ML/DL development and evaluation, with reliability, privacy, and security becoming practical tools when models move toward deployment.
I completed an MS by Research at IIIT Hyderabad under C. V. Jawahar and Anand Mishra, and previously worked as a research intern at Intel Labs in Bengaluru with Nataraj Jammalmadakka.