Visit my Portfolio! →

Web.

Web app @sushrutalgs.ai

Built the web application for sushrutalgs.ai, an AI study assistant for advanced surgical exam prep that answers questions with citations traced back to standard textbooks. It delivers a streaming chat interface with branching conversations, inline textbook figures and tables, and the marketing, sign-up, and onboarding flows.

Source is private; sushrutalgs.ai is a live product. Happy to walk through the code or grant read access on request.

Links

  • live product↗
  • Request repo access→

Stack

  • Next.js 16
  • React 19
  • TypeScript
  • Tailwind CSS v4
  • Supabase
  • Cloudflare R2
  • Vercel

System architecture. Tap to enlarge.

Overview

The web app is the primary sushrutalgs.ai client and marketing site. It delivers the streaming chat experience, the figure and table citations, and the sign-up and onboarding flows that bring a user into the product.

Approach

  • Branching conversations. Retry and edit create sibling branches in a tree-structured message model, driven by a reducer store and mirrored in Postgres so a conversation can be resumed anywhere.
  • Streaming UI. A server-sent-events client decodes tokens, thinking steps, citations, and figure and table artifacts as they arrive from the gateway.
  • Hardened data path. Auth-gated Cloudflare R2 proxies serve textbook figures and tables, behind a build-time content-security policy and redaction-wrapped scripts.

Results

About 27,300 lines of TypeScript across 51 components, 18 routes, and 6 API routes, with 23 consecutive Supabase migrations hardening row-level security, quota enforcement, and search. Clean production build with a roughly 2 MB client footprint.

Engineering

Next.js 16 and React 19 with Tailwind CSS v4, Supabase for auth and Postgres, Cloudflare R2 for assets, and Resend for email, deployed on Vercel. Model inference is delegated to the sushrutalgs.ai BFF, so the web client never holds the backend key.