Rohan Tikotekar

Trustworthy Autonomous Systems Lab, UC Riverside

Riverside, CA · Research

• Designed a multi-modal analytical framework fusing Vision-Language Models (VLMs) with spatial data features to predict anomaly-intent thresholds across complex variables (submitted to IROS 2026).
• Achieved 20% faster model convergence in LLMs vs. baseline statistical trends by integrating historical variance and feature-engineering data into the reasoning context.
• Mitigated operational risk by 30% with a strict confidence-modulated governance protocol that shares execution controls between automated models and manual validators.

Digital Vision Studios

Pune, India · Internship

• Engineered and delivered a high-concurrency ledger and secure transactional data system for 5,500+ active users under strict data-governance, integrity, and risk-mitigation standards.
• Orchestrated a migration to a hybrid cloud architecture (AWS / DigitalOcean), scaling capacity to process 8,000+ daily analytical requests.
• Automated workflows and built deployment guardrails with Docker, reducing delivery cycles from 2 hours to 15 minutes.

Atlas Copco

Pune, India · Internship

• Led the analysis, design, and delivery of an AI vision pipeline using YOLOv8 to validate manufacturing operations.
• Achieved 92.5% defect-detection accuracy while reducing per-unit inspection time from 45 seconds to under 2 seconds.
• Automated handwritten-checklist tracking with multimodal AI (image + text), reducing per-checklist processing from 5 minutes to under 1 minute (80% reduction).
• Built a Dockerized MLOps pipeline and partnered with operations teams to deliver a full-stack warehouse management system (Django & React.js).

Vishwakarma Institute of Information Technology (VIIT)

Pune, India · Research

• Conducted applied research on robust facial detection and recognition for occluded and profile faces using Convolutional Neural Networks and Haar Cascade classifiers.
• Engineered a deep-learning pipeline spanning image preprocessing, patch splitting, and EDSR-GAN super-resolution, improving detection accuracy to 93.2%.
• Published and presented the work at the Scopus-indexed 9th ICICT 2024, London (Springer Nature).