SRUTHI SRINIVASAN

• Realtime Face Detection, Face Recognition, Face anti-spoofing for HRMS applications in Zoho (Models used : SSD from torch vision, YOLO, MobileNet V3).Increased accuracy from 67% to 93% average in FAS.
• Actively contributed in dataset preparation in training/Fine-tuning Opensource VLM's like Deepseek, Qwen, InternVL.
• Currently working on developing a multimodal AI models with image and text encoders and structured-output decoders to extract data from documents and generate outputs like JSON (Models Used: Vision - Qwen 2.5 VL encoder 600M, Decoder : Qwen 2.5 500M)
• Developed Table structure recognition and layout model from scratch and achieved accuracy of 92%.
• Performed rigorous testing and validation of AI models to ensure robustness and reliability in real-world applications.
• Optimized machine learning model performance with hyperparameter tuning techniques, achieving improved accuracy and generalization capabilities.
• Mentored junior engineers on machine learning concepts and best practices, fostering a culture of continuous learning and collaboration.

• Experience with building end-to-end ML pipelines: data preprocessing, model training, evaluation, and deployment
• Developed a Sales forecasting engine using ensemble machine learning models like RandomForest, GLM,XGBoost and time series models like ARIMA and HoltWinters . Achieved accuracy of about 90% avg.
• Provided Key Insights to the Client by providing

Data Driven Insights : Sales & Trends, Forecast

Competitive Analysis analysis : SWOT Analysis

Marketing improvements : New Market exploration

Customer Segmentation : Behavioural Insights

Product & Improvement : Customer Feedback & R&D direction