Dhyanesh Vinothkumar
Software Development Intern
INFORIOS SOFTWARE TECHNOLOGIES INDIA PVT LTD
- Built DocFill, a parallel OCR ensemble (AWS Textract, PaddleOCR, Nova Vision) with a spatial consensus engine that removes character-level errors
- Reached 98%+ field accuracy (PAN, Mobile, Pincode) and a +25% confidence boost, with ~12 s end-to-end AWS Lambda latency
- Built a company-wide Project Tracker (React/TS, GraphQL, AWS Lambda, DynamoDB) to replace WhatsApp-based project updates across teams
- Led DynamoDB schema design, cross-tenant isolation, and CI/CD pipeline setup
Production OCR ensemble that runs three engines in parallel and fuses their outputs with a spatial consensus engine. 98%+ field accuracy on PAN, Mobile and Pincode, with ~12 s Lambda latency.
Production OCR ensemble that runs three engines in parallel and fuses their outputs with a spatial consensus engine. 98%+ field accuracy on PAN, Mobile and Pincode, with ~12 s Lambda latency.
Self-updating engine that scrapes 27+ sources across 8 regions for internships and startup opportunities. Keeps 85+ live entries, re-indexes every 24 hours, and sends Discord/Slack alerts.
End-to-end ML pipeline with a 7-model benchmark and GridSearchCV tuning. Best model was Ridge Regression (R² = 0.88). Deployed as a Flask API on Azure with CI/CD. Tech used: Python, scikit-learn, XGBoost, CatBoost, Flask, Azure, GitHub Actions
4-class heart disease severity classifier on 1,337 patient records, using SMOTE for class imbalance. Best of 9 models was Gradient Boosting (76.03% accuracy, F1 = 0.7659).
Btech Computer Science and Engineering
VIT-AP · 8.55
Attendee · Oct 24, 2026