IQAILAB Private Limited is a deep-tech AI research lab building next-generation autonomous learning systems, computer vision models, and scalable AI infrastructure.
Developing self-improving reinforcement learning agents capable of multi-task generalization without explicit reward engineering.
High-efficiency object detection and segmentation models optimized for edge deployment with sub-10ms inference latency.
Building distributed training pipelines and model serving infrastructure for cost-efficient large-scale AI experimentation.
Our research programs are structured around three core pillars—each driving systematic advancement in artificial intelligence methodologies.
Investigating meta-reinforcement learning paradigms where agents develop intrinsic motivation signals. Our approach combines model-based planning with curiosity-driven exploration to achieve sample-efficient generalization across diverse task distributions.
Developing lightweight vision transformers and efficient convolution architectures for real-time object detection on resource-constrained hardware. Our models achieve state-of-the-art accuracy with 4x fewer parameters.
Designing distributed training frameworks with dynamic resource allocation, automated hyperparameter optimization, and fault-tolerant model checkpointing for cost-efficient experimentation at scale.
A focused founding team driving AI research from hypothesis to deployment.
Founding CEO of IQAILAB Private Limited. Leads strategic direction, research agenda, and organizational development. Focused on building a sustainable AI R&D operation that bridges fundamental research with real-world deployment.
Co-founder and technical research lead. Drives hands-on AI experimentation across autonomous learning, computer vision, and infrastructure optimization. Responsible for experimental design, model development, and technical publication output.
Registered as IQAILAB Private Limited. An Indian deep-tech company focused exclusively on AI research and experimental development.
To advance artificial intelligence through rigorous, reproducible research—building scalable systems that transition from experimental prototypes to production-ready solutions.
Hypothesis-driven experimentation with transparent benchmarking, open publication of findings, and a commitment to efficient, cost-conscious AI development.
Technical papers, experimental reports, and methodology documentation from our ongoing research programs.
S. Oraon, N. Kumari — IQAILAB Technical Report
We propose a novel framework combining curiosity-driven exploration with model-based planning, achieving 94.7% task completion across 48 unseen environments with 12x improvement in sample efficiency.
N. Kumari, S. Oraon — IQAILAB Technical Report
A knowledge distillation approach for compressing large vision transformers into sub-10M parameter models while maintaining 91.2 mAP on COCO, enabling deployment on NVIDIA Jetson and mobile platforms.
S. Oraon, N. Kumari — IQAILAB Technical Report
An elastic scheduling system for multi-GPU training that dynamically reallocates compute based on gradient noise ratios, achieving 3.8x throughput gains with 62% cost reduction on cloud infrastructure.
N. Kumari, S. Oraon — IQAILAB Technical Report
An open benchmark framework standardizing evaluation metrics across reinforcement learning, computer vision, and NLP tasks, with automated reporting and comparison against published baselines.
Interested in collaboration, partnerships, or joining our research team? Reach out below.