AaryanSharma.

Building GCP-native ML infrastructure for cricket analytics — highlights, biomechanics, and coach workflow automation. Shipped a production full-stack platform (30+ endpoints, 35+ database models, 25K+ LoC) designed for scale: Cloud Run autoscaling, async job processing, signed-URL uploads, and role-based access control.

• →Built a scalable cricket analytics pipeline on Google Cloud Storage, Cloud Run, and Cloud Tasks to asynchronously process 7+ hour match videos, eliminating frontend timeout failures and improving processing throughput by 3x.
• →Engineered an OCR-based highlight generation system using frame sampling, rolling-median smoothing, and cricket event classification, reducing coach analysis time from 7 hours to 20 minutes per match at 90% detection accuracy.
• →Developed a MediaPipe Pose-based biomechanics engine for batting and bowling, extracting 15+ performance metrics, identifying technical flaws, and generating AI-assisted coaching recommendations and PDF reports.

Built Navixar end-to-end — a SaaS compliance platform that turns form responses into AI-generated regulatory reports — owning the frontend, backend, and AI evaluation pipeline.

• →Designed a 4-layer prompt-orchestration engine for Gemini 2.5 Flash (facts / global rules / session memory / scoped synthesis) to eliminate generic and hallucinated output, replacing naive single-shot prompting
• →Built a non-blocking evaluation engine using threaded background execution so form submissions return instantly instead of blocking on high-latency AI calls, with per-request token and cost logging
• →Implemented hybrid JWT authentication with token blacklisting and full account lifecycle (signup, email verification, password reset, deactivation, deletion), plus a credit-based rate-limiting system to meter AI usage per client

Engineered ESP32 hardware interface libraries for Smowcode's visual IDE

• →Developed and validated ESP32 peripheral libraries using ESP-IDF, ensuring memory safety and thread-safe execution for real-world deployments
• →Engineered C integration layers linking ESP-IDF hardware libraries to a visual IDE, utilizing JSON configuration files to map hardware pins and manage library dependencies
• →Implemented ADC, DAC, and RF calibration with testing and validation

Selected as a Google Student Ambassador to represent Google at KJSCE, organize events, and promote Google technologies on campus.

• →Organized and led technical workshops and meetups
• →Promoted Google technologies and initiatives on campus

Led a 7-member creative team to define branding, visual systems, and event identity across ISTE KJSSE initiatives.

• →Delivered graphic and motion content for campus technical events
• →Coordinated cross-team assets with PR and tech leads to boost event turnout

Designed scalable visual assets and reusable branding systems for ACM events, outreach, and chapter communications.

• →Created 35+ visual assets for events and outreach
• →Achieved 40% growth in social media reach