Looking for a job

Abhishek Agrawal

@abhishekagrawal3

Message

I build production computer vision and vision-language systems that improve accuracy, latency, and cost at scale.

India

Message

What I'm looking for

I'm looking to lead hands-on machine learning and computer vision work that takes foundation models from data curation and experimentation to reliable, scalable production systems.

At EagleView, I lead end-to-end computer vision initiatives spanning vision foundation models, multimodal embeddings, temporal change detection, line detection, and image segmentation.

I've developed vision foundation models on 10M images, reducing fine-tuning data requirements by over 50%, and deployed capture-anomaly detection at 95% accuracy. My LoRA-fine-tuned vision-language embedding model improved Recall@1 from 0.03% to over 50% for multimodal retrieval.

I architected instance segmentation for 80+ classes, reducing operational costs by 67%, and built AWS-based ML microservices for 1.26M+ annual inference requests. I also optimize PyTorch inference through ONNX, TensorRT, and NVIDIA Triton.

Previously at Blaize and TheHiTechRoboticSystemz, I worked on quantization tooling, fixed-point math, ADAS multi-object tracking, object detection, depth estimation, and traffic-light classification.

Experience

Work history, roles, and key accomplishments

Current

Lead Data Scientist

Current

EagleView

Feb 2022 - Present (4 years 7 months)

Drive multiple end-to-end Computer Vision initiatives, including Vision Foundation Models, Multimodal Embeddings, Temporal Change Detection, Line Detection, and Image Segmentation. Led development of vision foundation models on 10M images, reducing fine-tuning data requirements over 50%.

Computer Vision Deep Learning Foundation Models Multimodal Learning Self Supervised Learning CLIP LoRAs Onnx TensorRT NVIDIA Triton Inference Server AWS Sagemaker AWS Lambda AWS Step Functions PyTorch Python

Research Engineer II

Blaize

Feb 2019 - Feb 2022 (3 years)

Developed tooling to reconstruct PyTorch computational graphs from ONNX models, enabling INT8 and INT16 post-training quantization. Explored LoRA through tensor decomposition for transforming 3D CNNs to 1D CNNs.

PyTorch Onnx Quantization Tensor Factorization Fixed Point Arithmetic C++

Research Engineer

TheHiTechRoboticSystemz

Jun 2017 - Jan 2019 (1 year 7 months)

Implemented Kalman Filter-based multi-object tracking for ADAS systems, improving tracking stability for vehicles and pedestrians for Time to Collision Estimation. Collaborated on projects involving Road Scene Object Detection, Unsupervised Monocular Depth Estimation, and Traffic Light Classification.

Kalman Filter Multi Object Tracking ADASエンジニア Object Detection Depth Estimation Computer Vision Python C++