Varun Khare

CEO

San Francisco, California, United States8 yrs 6 mos experience

Most Likely To SwitchAI Enabled

Key Highlights

Pioneered on-device AI for real-time applications.
Led global teams in federated learning research.
Developed open-source tools for AI-native applications.

Stackforce AI infers this person is a SaaS expert specializing in on-device AI and federated learning.

Contact

varun.khare@nimbleedgehq.ai LinkedIn

Skills

Core Skills

On-device AiAi InfrastructureFederated LearningOn-device Federated LearningMeta-learningNeuroscience ApplicationsDeep Learning3d ModelingAndroid Development

Other Skills

3D Computer Vision3D Segmentation3D U-NetAIAI Infrastructure & KernelsArtificial IntelligenceBayesian methodsBuilding Developer Platforms & EcosystemsC++Cognitive Behavioral PatternsComputer VisionData AnalysisDecentralized Inference & TrainingDecentralized SystemsDeepLab v3

About

I’m the Founder & CEO of NimbleEdge, where we’re reimagining how AI is delivered: locally, privately and in real-time, right on your device. Check out our platform open sourced at github.com/NimbleEdge/deliteAI NimbleEdge is pioneering on-device, session-aware AI to power smarter, more responsive mobile and applications without sending data to the cloud. From quantised LLMs like LLaMA 3.2 to streamlined speech pipelines, our stack makes it possible to run AI models entirely offline with zero compromise on performance or privacy. With roots in AI research (UC Berkeley, Max Planck) and on-device deployment, I bridge the gap between cutting-edge science and real-world productisation. My mission is to make AI real-time, private and omnipresent (from smartphones to wearables to sensors) Our open-source contributions (e.g., tokenizer batching, Sparse Transformers, Kokoro TTS) are helping developers across the world build AI-native applications that respect user autonomy and data dignity. I focus on - On-Device AI and ML - Decentralized Inference & Training - AI Infrastructure & Kernels - Building Developer Platforms & Ecosystems If you're solving for low-latency AI, in-device decisioning, or building apps that need local intelligence, let’s connect.

Experience

Nimbleedge

CEO & Co-Founder

Mar 2021 – Present · 5 yrs · San Francisco Bay Area · On-site

Building NimbleEdge. Turning cutting-edge AI research into real-world app experiences. Scaling AI directly on-device, because your smartphone is smarter than the cloud.

On-Device AI and MLDecentralized Inference & TrainingAI Infrastructure & KernelsBuilding Developer Platforms & EcosystemsOn-Device AIAI Infrastructure

Openmined

2 roles

Federated Learning Lead

Nov 2020 – Dec 2021 · 1 yr 1 mo

◦ Leading a multi-national team of 10 consisting of research scientists and engineers.
◦ Vetting research projects in federated learning for deployment into OpenMined stack.
◦ Devising novel algorithms for privacy preserving optimization and aggregation in federated
settings.
◦ Developing decentralized FL implementations and privacy-accuracy trade-offs for FL

Federated LearningPrivacy-Preserving AlgorithmsDecentralized Systems

Core Developer

Jan 2020 – Nov 2020 · 10 mos

Responsible for the development of secure ML algorithms on android using PySyft.
Funded by PyTorch and the RAAIS Foundation.
◦ Built & led the world’s first open-source ecosystem for on-device federated learning.
◦ Focussed on developing and maintaining open-source libraries such as PyGrid and KotlinSyft.
◦ Added support for peer-to-peer communication, secure aggregation, and SMPC protocols.
◦ Helped grow Open Source contributions to the community and presented the work at PriCon 2020

Secure ML AlgorithmsOpen-Source DevelopmentAndroid DevelopmentOn-Device Federated Learning

University of california, berkeley

Visiting Research Scholar

Jun 2020 – Nov 2020 · 5 mos · United States

Worked with Prof Dawn Song on Automating Program Synthesis via meta-learning, targeting generalization across diverse tasks and domains.
Key highlights of the work include:
◦ Implemented and evaluated methods like MAML (Model-Agnostic Meta-Learning) and transfer learning for SQL generation and task-specific adaptations.
◦ Designed hierarchical model architectures, integrating global transformer layers, LSTM cells, and domain-specific multi-headed attention mechanisms.
◦ Improved token prediction accuracy by optimizing data utilization in structure prediction and token matching.

Meta-LearningProgram SynthesisMachine Learning

Max planck institute for brain research, frankfurt am main

Visiting Research Scholar

Aug 2019 – Mar 2020 · 7 mos · Frankfurt Am Main Area, Germany

Trying to decipher cognitive behavioral patterns that can be implemented in AI while advancing neuroscience EM data analysis with Prof Moritz Helmstaedter
◦ Focused on automating myelin segmentation in 3D mSEM (multi-Scanning Electron Microscopy) data for large-scale connectomic analysis.
◦ Developed a 3D U-Net and DeepLab v3 hybrid model, achieving over 97% accuracy in 3D axon segmentation.
◦ Implemented a distributed pipeline to skeletonize 3D segmentation masks into connected components.
Pioneered the first successful automation of Peta-Byte scale axon detection in brain tissue, enabling practical analysis of massive mSEM datasets.

Cognitive Behavioral Patterns3D SegmentationDeep LearningNeuroscience Applications

National university of singapore

Visiting Research Scholar

May 2018 – Jul 2018 · 2 mos · Next++ centre

Worked with Prof Tat Seng Chua on 3D computer vision generating 3D models from single RGB image.
◦ Implemented monocular SLAM for generating 3D point clouds and tested the approach using Kinect.
◦ Proposed and initiated a long-term project on texture synthesis for unseen 3D object poses, requiring the extraction of canonical object poses.
◦ Implemented MarrNet and Faster RCNN in PyTorch.
◦ Contributed to open-source projects in generative voxel modeling for creating 3D voxels.

3D Computer VisionMonocular SLAMGenerative Modeling3D Modeling