Work Experience
- Driving innovation in memory-centric solutions for LLMs, advancing AI applications to enhance human productivity and efficiency.
- Implemented a Neo4j-based graph-memory for efficient entity storage and retrieval, boosting context awareness and response accuracy.
- Enhancing image quality globally and locally, focusing on selective detail enhancement and comprehensive filtering via deep learning.
- Developed a deep learning network for detecting similar HDR bracket images, achieving artifact-free merging.
- Engineering automated censoring for license plates, faces, and other sensitive data, ensuring image compliance with data protection laws.
- Utilized prompt engineering and memory-based approaches in large language models for text-based game playing.
- Designed reinforcement learning agents with enhanced memory, language grounding, and object affordances. (Published at KCap 2023)
- Enhanced visual question answering system by implementing visual cropping methods to focus specific region in images. (Published at NeurIPS Workshop 2023)
- Developed a multimodal approach for recipe generation from food images using attention-based vision and language models. (Published at WACV 2024)
- Utilized analytical and technical expertise to deliver insights and suggestions for business progress.
- Achieved 92% precision on fraud engine by leveraging historical and session-level attributes to block fake leads.
- Developed a computer vision pipeline creating 3D floor plans from spherical projections, saving 400 work hours/month.
- Automated image auditing process, saving $40k/year and increasing online traffic by 12%.
- Devised an image classifier with 93% precision and 89% recall, labeling real-estate tags in real-time at 6 FPS on CPU. (Article)
- Published two research papers in top international conferences: ECML'22 and CODS-COMAD'23
- Improved image quality (contrast, brightness, sharpness, blur) resulting in 8% increase in CTR on social media campaigns. (Article)
- Developed a No-Reference Image Quality Ranking tool to select the most visually appealing real estate property images. (Article)
- Launched a service that extracts real-estate entity tags from WhatsApp group messages with 87% f1-score.
