Impact & Achievements
Here’s a snapshot of what I achieved during my AI/ML Internship at SynerSense Pvt. Ltd. from May 2025 onward.
Key Statistics
| Metric | Value |
|---|---|
| Weeks Completed | 11+ |
| Tools Mastered | 10+ |
| VLM Prompts Created | 15+ |
| Research Papers Studied | 7 |
| Custom Gradio Apps Built | 3 |
| Models Evaluated | 5+ |
| Score Consistency Experiments Run | 100+ evaluations |
| Final Prompt Versions Optimized | v9 and v10 |
Technical Milestones
- Built end-to-end image-based scoring interfaces for facial region quantification
- Created prompt tuning iterations using reasoning-based formatting and region-wise control
- Prototyped multi-model benchmarking across deployment endpoints (OpenAI, HF Spaces)
- Researched and implemented score caching and visual consistency strategies
- Designed prompt formats to improve float stability using few-shot and no-shot methods
- Explored Chain-of-Thought prompting and dataset curation strategies
Tools Used
- Languages/Frameworks: Python, PyTorch, Jupyter, Markdown
- Libraries: OpenAI SDK, Gradio, ComfyUI, Hugging Face Transformers
- Platforms: Hugging Face Spaces, GitHub, OpenAI Playground, Replicate
Highlights
- Improved float scoring consistency by refining formatting and input structure
- Designed an experiment dashboard for input-output variance testing
- Drafted micro-research blogs analyzing prompt determinism and visual scoring challenges
- Achieved ~25–30% improvement in prompt reliability over baseline formats
Continuous Learning
Every experiment was backed by reading, testing, and documenting—ensuring reproducibility and contributing to future toolkits in multi-modal AI inference pipelines.