TABISH SHAIKH

• Led a team of 5 engineers to design and deliver a real-time AI safety pipeline for detecting violence, NSFW, and harmful content,
achieving 92.72% classification accuracy on real-world data.
• Architected a dual-model system (EfficientNet + motion-based model) for image and video analysis, enabling low-latency, on-device
inference using TFLite.
• Built and managed end-to-end dataset pipelines (50k+ samples), including data collection, annotation, cleaning, and quality validation for
multi-class content classification.
• Developed and deployed an Android-integrated AI monitoring system for continuous background scanning with a privacy-focused, offline
first architecture.
• Defined system architecture, model strategy, and execution workflows, improving team delivery efficiency and pipeline scalability.

• Developed and maintained production-grade full-stack applications including e-commerce, AI utility, and educational management platforms
using React, Node.js, and PostgreSQL.
• Built and optimized backend APIs and core business logic for Begorgeous.in, handling product catalog, user workflows, and scalable
order management systems.
• Engineered Tryatlabs.com, a scalable AI tools & utility ecosystem platform with modular React + Vite architecture, SEO optimization,
and AdSense-ready monetization structure.
• Contributed to the development of a Student Intelligent System (SIS) platform for educational workflow management, including role-based
modules, dashboards, and scalable data handling systems.

▸ Engineered a real-time AI content moderation system in Python, achieving 87.72% classification accuracy with sub-200ms
inference latency for production-scale traffic.
▸ Developed a real-time Audio AI / Speech Processing model that reduced pipeline latency by 37% over baseline, enabling near-real
time conversational AI responses.
▸ Fine-tuned a YOLOv8-based object detection model on 20K+ annotated images, achieving 0.80+ mAP for workplace safety
violation and restricted-zone detection.
▸ Designed a scalable document chunking and retrieval system (RAG-ready) that improved downstream AI task efficiency by 37%
and reduced document processing time significantly.
▸ Built an OCR + NLP document parsing pipeline using Tesseract and transformer-based models, achieving 79.28%+ field extraction
accuracy on structured enterprise documents.
▸ Owned end-to-end ML pipeline engineering — dataset versioning, preprocessing, model training, evaluation, and CI/CD-style
deployment workflows.