I will train a vision language model for any grounding task

Do you need a Vision-Language Model that works 

on YOUR data not generic benchmarks?

VLMs are powerful. But out of the box, they fail 

on custom domains. The language grounding breaks 

down. The visual attention scatters. Performance 

collapses sometimes to near-random levels.

I fine-tune vision-language models for custom 

tracking, grounding, detection, and retrieval 

tasks across & domain you are working in.

What I have worked on:

• - Language-guided multi-object tracking in surveillance & urban environments
• - Domain adaptation from traffic datasets to fixed-camera real-world scenarios
• - Reduced manual labor of annotations for large datasets upto 70 % by engineering automated piplines.
• - Made benchmarks for custom niche use case .

What I deliver:

- Fine-tuned model trained on your dataset

- Performance evaluation and metrics report

- Explainability outputs on request

- Clean, documented code fully yours

I work with transformer-based architectures, 

modality fusion in VLMs, and custom annotation 

pipelines. I understand where these models 

break and how to fix it.

Message me first with your use case and 

dataset I will tell you exactly what 

is achievable before you commit.