6 hours ago
Bengaluru, IndiaMid Level
Responsibilities
- Develop image-generation, virtual try-on, identity-preserving, and reference-based generative AI models.
- Train and fine-tune models using PyTorch, LoRA, personalization, and adapter-based techniques.
- Prepare datasets, build training pipelines, and evaluate model quality and performance.
- Convert and deploy models using ONNX, ONNX Runtime, TensorRT, Core ML, and TFLite/LiteRT.
- Apply FP16, INT8, INT4, pruning, compression, distillation, and other optimization techniques.
- Profile and debug inference across CPU, GPU, NPU, and Apple Neural Engine hardware.
- Reduce latency, peak memory usage, battery and thermal impact, and model footprint.
- Build production-ready pipelines spanning data, training, evaluation, conversion, runtime optimization, on-device deployment, and monitoring.
Requirements
- 2–5 years of experience.
- Strong experience with Python, PyTorch, and deep learning.
- Hands-on experience with computer vision and generative AI.
- Experience training or fine-tuning image-generation models.
- Good understanding of LoRA, model optimization, and quantization.
- Strong knowledge of inference runtimes such as ONNX, TensorRT, Core ML, TFLite/LiteRT, or ONNX Runtime.
- Strong debugging, profiling, and software engineering fundamentals.
- Preferred experience with FLUX, Stable Diffusion, SDXL, ControlNet, IP-Adapter, VAE, CLIP, CUDA/GPU optimization, Metal, Apple Neural Engine, Android NNAPI, GPU delegates, C++, Swift, Kotlin, distributed or multi-GPU training, model compression, pruning, and knowledge distillation.
About Glance
Glance builds AI-driven commerce and content platforms that surface personalized shopping and media on mobile lock screens, connected TVs, and apps. Its products include Glance, Roposo, and Nostra, partnering with smartphone and TV manufacturers, telecom providers, and brands to enable discovery-led shopping and entertainment, monetized through advertising and commerce. Founded in 2019 and headquartered in Bangalore, it operates as an unconsolidated subsidiary of InMobi, is backed by Google and Jio Platforms, and runs on 450M+ devices.