NLP & Image Processing Engineer

WorkSprout is a deep-tech engineering and design partner across twelve practice areas — TinyML & Edge AI, Custom AI Development, AI Automation & Agents, Data Engineering & Collection, IoT Solutions, Hardware Prototyping, Robotics & Automation, 3D Design & Modeling, Software Development, Growth & DevOps, Branding & Creative Design, and Startup & Product Launch. We deliver systems that work in production, not just in demos.

You will develop real-world NLP and image processing solutions — from OCR pipelines and document intelligence to multi-modal AI systems that understand both text and visual content in production environments.

• Build and fine-tune NLP models for text classification, extraction, and summarisation
• Develop image processing pipelines using OpenCV and deep learning frameworks
• Implement multi-modal models combining text and vision understanding
• Create OCR and document intelligence systems for structured data extraction
• Deploy models as low-latency microservices with REST interfaces
• Evaluate model performance against real-world accuracy and recall benchmarks

• Strong Python with PyTorch or TensorFlow
• Experience with Hugging Face Transformers and pre-trained NLP models
• Proficiency in OpenCV and image processing fundamentals
• Understanding of CNN, ViT, and multi-modal architectures
• Experience deploying ML models to production (ONNX, TorchServe, FastAPI)

• Experience with Bengali or multilingual NLP pipelines
• Knowledge of Tesseract OCR and document layout analysis
• Familiarity with CLIP, BLIP, or LLaVA for vision-language tasks

• Work on diverse real-world language and vision problems
• Competitive salary benchmarked to international rates
• Flexible hybrid work arrangement
• GPU compute access for training and experimentation
• Dedicated research time for paper reading and prototyping

* Benefits marked with an asterisk apply to permanent employees only.

If your experience matches the requirements and you are ready to work on production engineering and design challenges, submit your application using the button below. We review every submission personally — no automated screening, no ghosting.