
Serve as the principal architect, spearheading the development and design of Dylog Inc's conversational AI and search platform.
- Developed pipelines for synthetic data generation, tailored for e-commerce platforms, improving search results by 20% (internal evaluations).
- Developed a hybrid search framework combining Elasticsearch, ColBERT, and llama, integrating keyword-based and vector-based retrieval for enhanced search accuracy.
- Designed and implemented agentic workflows for automating sales data processing from emails, text messages, and contact forms, reducing order processing time by 2 hours.

Working with Professor Hao Zhang on multi-modal learning approaches.
- Experimenting with multi-modal learning approaches based on DETR and GLIP to obtain attribute-based representations from natural language descriptions and visual features for object instance recognition and tracking.

Implemented advanced machine learning solutions for user behavior analysis.
- Implemented a multi-modal deep learning model for categorizing of user behavior based on VoIP traffic, images, and textual data gathered from user browsing sessions.
- Worked end-to-end on the project, from data collection to successful deployment.

Worked with Data Optimization and Backup team on database solutions.
- Developed an algorithm to improve and optimize database re-shading method in Eon mode.
- Extended this algorithm to improve backtracking of newly created storage containers while performing operations like re-shard, ILM and alter table.
- Designed and developed pipelines for data analytic and machine learning teams to work on backup and restoring information.

Research on cross-modal transformer approaches for video analysis.
- Proposed an advanced cross-modal transformer approach with a video frame-level contrastive loss to reveal the correlation between language queries and video clips, focusing on Natural Language Query tasks for the Ego4d dataset.
- This approach resulted in an improvement in the baseline with a IoU 0.5 recall @ 5 of 8.2 and recall @ 3 of 4.8.

Lead architect, developer, and designer for RenderPub Studio, Stitch & Walk, architectural visualization softwares.
- Developed Common Unreal Modules (CUM) - a modular C++ / Blueprint system for RenderPub's Unreal Engine based applications.
- Developed and implemented a 360-degree panoramic image loading algorithm utilizing tile-based streaming, three.js, WebGL, and OpenCV.
- Developed a desktop application utilizing Next.js, Electron, and React to launch and auto-update the RenderPub suite of applications and projects.
Yourkraft, a marketplace connecting entertainers and party planners, secured funding from renowned singer Vijay Prakash.
- Managed a network of over 200 artists across multiple cities, including Bangalore, Mumbai, and Hyderabad, and successfully hosted over 350 events.
- Developed a feature-rich dashboard for event planning and artist hiring, including an events list and the ability to book event tickets, utilizing Flask and Firebase technology stack.
Vayam is a student-led trust to empower individuals through education and equality as its foundation.
- In partnership with five government schools around Mysore, providing online classes in subjects such as science, mathematics, and computer science to approximately 200 children.
- Organizing various community initiatives, including blood donation drives, charity events, food relief efforts, and green initiatives.

Focused on advanced computer science topics with emphasis on AI and machine learning.
- Coursework: Neural Networks, Advanced Natural Language Processing, Intelligent Visual Computing, Mathematical Statistics, Systems for Data Science, Reinforcement Learning

Fundamental computer science education with focus on practical implementation.
- Coursework: Operating System, Data Structures & Algorithms, Data mining and warehousing