Durgesh Dutt Sinha

RoleRadar is an enterprise-grade career intelligence platform evaluating candidate resumes across 16 modern tech roles with an automated 8-point ATS compatibility scanner, dynamic Google XYZ bullet optimizer, and personalized multi-week learning roadmaps.

• User uploads resume document (.pdf / .docx) or pastes text directly.
• Document parsing pipeline extracts text, sanitizes non-standard Unicode, and validates byte limits.
• Keyword normalizer matches terms against the 16-role benchmark or user-supplied custom Job Description.

A production-grade 9-page multi-modal fitness architecture utilizing Google MediaPipe 33-point skeletal landmark detection, trigonometric joint-angle vector kinematics, and web worker isolation to provide real-time form correction under 50ms latency.

• User grants webcam permission; video element acquires 720p feed.
• Frames are captured and piped into a dedicated Web Worker via transferable ImageBitmap.
• MediaPipe predicts 33 3D skeletal coordinates with sub-50ms inference.

An enterprise customer segmentation and lookalike audience platform transforming 4,000 retail records into actionable marketing cohorts using log-transformed RFM feature engineering, K-Means ($K=5$), Gaussian Mixture Models (GMM), and DBSCAN anomaly detection.

• Raw transaction ledger ingested via CSV (Customer ID, Invoice Date, Quantity, Price).
• RFM feature transformation computes Recency, Frequency, and Monetary Value per customer.
• Log1p scaling and StandardScaler normalize skewed monetary distributions.

• Engineered end-to-end AI deployment pipelines with automated containerization.
• Constructed and benchmarked autonomous agent workflows for enterprise industry use-cases.
• Executed production-grade machine learning models under direct technical mentorship.

• Developed multi-step LLM prompt chaining architectures to automate repetitive tasks.
• Built full-stack AI concept prototypes from initial system design to live client demonstrations.