* Developed Image Forgery Detection (IFD) and PDF Forgery Detection products for major clients like BofA, MetaMap, AXA, and Uber.* Engineered advanced document skew correction (Page Dewarp) and integrated PaddleOCR, boosting OCR accuracy by 10% and 25% respectively for various financial documents.* Implemented Image Quality Assessment (IQA) for enhancing real-world documents and improving OCR accuracy on low-quality images.* Designed and maintained deep learning models for Image Forgery Detection, Automatic Signature Cleaning, and Verification.* Managed the complete ML lifecycle for multiple projects, including data collection, generation, exploration, and training.* Text enrichment which increased LLM extraction accuracy by 15%.* Spearheaded the development of Tabular Framework, an LLM-powered product for detecting and extracting tabular data from various document types by accepting natural language prompts.* Created benchmark datasets and scoring modules to ensure continuous product improvement and early regression detection.* Contributed to building RAG pipelines for long documents and Agentic workflows for complex, multi-step tasks.* Fine-tuned open-source SLMs using LoRA and constructed an Enterprise DB Search component using Weaviate Vector DB and LLMs.* Led hackathon teams, resulting in new product features, and participated in hiring processes, including giving pre-placement talks and mentoring interns on Image Forgery Detection and Semantic Document Comparison projects.

• Individual contributor (95%)• Mentoring a team of 12-15 engineers (5%)• C++ Product Development• Utilized ML algorithms like SVM and Gradient Boosted Machines to achieve Cross Validated performance of more than 95%, using domain specific features• Achieved 98% character classification score using Fully Connected Neural Network architecture of Deep Learning without handcrafted features.• Created Convolutional Neural Network (CNN) model for offline handwriting recognition which improved results by 20%

• Analyzed large data sets and developed statistical and classification models using Kernel Density Estimation, Logistic Regression and Random Forests which improved accuracy by 6%• Built Ensemble Models using Stratified Sampling which beat individual models built without stratification• Designed and implemented Handwritten Note-Taking Apps for Android, iOS and Windows as a front-end to the Recognition Engine and for data gathering.

• Implemented various Proof of Concepts in Matlab and C++• Feature Engineering using signal processing, curve matching and segmentation of data enabled performance to increase by 10%• Performed Feature Importance and Selection tests which resulted in improved models and faster training• Participated in all phases of data exploration: data aggregation, cleaning, validation, visualization, partitioning, augmentation and imbalance reduction• Streamlined Data Collection and Preparation phase.