Description :

We are hiring ML engineers focussed on testing, tuning and adapting publicly available computer vision models for our use-cases.

Youll work with modern open-source CV architectures and improve them via fine-tuning approaches like LoRA, hyper parameter optimization, evaluation harnesses and deployment-oriented experimentation.

Key Responsibilities :

- Evaluate and benchmark publicly available CV models (detection/segmentation/classification/) on internal datasets and metrics.

- Design, modify, and extend neural network architectures (e.g., adding/removing layers, modifying backbones, introducing new heads or attention modules) to support experimentation and performance improvements.

- Fine-tune models using parameter-efficient techniques such as LoRA/QLoRA/adapters and experiment with training strategies (freezing layers, schedulers, augmentation etc).

- Perform hyper-parameter tuning (learning-rate, batch size, optimizer/scheduler, regularization, augmentation knobs, LoRA, rank/alpha/dropout etc) to maintain experiment tracking.

- Build repeatable pipelines for :

a. Dataset preparation and labeling quality checks.

b. Train/val/test splits and ablation studies.

c. Metrics reporting and regression testing.

- Debug training issues (instability, overfitting, poor generalization, data leakage) and propose practical solutions.

- Optimize inference for cost/latency (quantization, batching, pruning as applicable) and help package models for production use.

- Collaborate with product/engineering to translate requirements into measurable offline + online performance outcomes.

Must Have Qualifications :

- Strong Python skills and hands on experience with PyTorch or Tensorflow/JAX.

- Practical experience working with computer vision models and training/fine-tuning workflows.

- Experience with LoRA/parameter efficient fine-tuning or comparable approaches.

- Solid understanding of training fundamentals : loss functions, optimization, regularization, augmentation, evaluation methodologies.

- Experience running experiments on GPUs (local and cloud) familiarity with tooling such as Weights and Biases/ ML Flow for tracking.

Good To Have Qualifications :

- Experience with HuggingFace ecosystem (Transformers, Diffusers, Datasets) OpenMMLab, Detectron2, Ultralytics etc.

- Experience tuning vision-language models or generative vision models (CLIP-like, Grounding, SAM-Style segmentation, diffusion fine-tuning).

- Knowledge of hyperparameter optimization frameworks (Optuna, Ray, Tune).

- Production exposure : model packaging, monitoring, CI for ML, data versioning, deployment (Triton, Torch Serve).

- Experience with quantization (bits and bytes, AWQ/GPTQ-Style for VLM stacks, TensorRT, ONNX).