Hi, I'm Sahil Raj

Transcripts from multi-hour calls are verbose and hard to digest. Standard LLMs often produce inconsistent schema outputs when queried for JSON structures, causing client-side rendering parser errors.

• JSON-mode model endpoints with strict schema validations.
• Custom TypeScript schema normalizer to map varied transcript structures.
• Interactive timeline and checklist views for items.

Guarantees structured API outputs, reducing post-processing error rates to near zero, and saving teams hours of manual review.

Extracting garment crops historically required uploading image catalogs to remote GPU-backed ML clusters, incurring significant network latency and high cloud compute bills.

• PoseNet model loaded locally inside the web browser.
• HTML5 canvas region rendering and crop math.
• Zero server dependencies allowing offline processing.

Eliminated backend hosting costs for ML nodes and reduced crop extraction latency to milliseconds, delivering an instant UI experience.

Managing compliance verifications and role permissions for Billing Houses, Providers, and Admins was manual, slow, and highly vulnerable to security gaps.

• Secure role-based access control (RBAC) schemas.
• Dynamically orchestrated multi-step validation forms.
• Full compliance audit logs tracking provider changes.

Successfully onboarded and secured records for 50K+ active users, cutting verification processing times significantly.

Rendering hundreds of shift cards on a single page with custom overlaps caused DOM bloat, heavy script execution, and laggy drag-and-drop operations.

• Virtualized list rendering keeping active DOM nodes low.
• Optimized Redux store updates reducing unnecessary redraw cycles.
• Drag-and-drop actions with collision validation.

Boosted client-side interaction performance by 40%, serving over 100K+ concurrent users with zero UI lag.