AI product interfaces · India / remote
Engineering case studies
Projects you can inspect
Source, implementation choices and current limits — for a practical review of the work.
These case studies explain the problem, implementation and current limits. Public source and live links let you assess the engineering directly.
Product application · Next.js / TypeScript / Vercel AI SDK
MK QuizFlow — study material to practice
Problem
Make pasted notes and text-based PDFs usable as quizzes and flashcards, including when AI credentials are unavailable.
Implementation
A Next.js application with browser-side PDF extraction, IndexedDB persistence, editable questions, keyboard controls and JSON, CSV and Markdown exports. Deterministic Quick mode provides a usable baseline; optional AI routes use the Vercel AI SDK and explicit unavailable states.
Engineering tradeoff
Local-first storage removes signup friction but does not provide cross-device account sync. Scanned PDFs require OCR, which the documented implementation does not yet include. Quick mode uses heuristics and is clearly labelled.
Try QuizFlow · Inspect source, tests and documentation
Developer tooling · Agent instructions / reusable skills
AI Assistants Ecosystem — repeatable coding workflows
Problem
Keep AI coding instructions and workflow knowledge organized across different developer assistants.
Implementation
A repository of assistant configurations, reusable skills and workflow documentation, supported by a public showcase. The OpenCode repository provides another entry point into the tool-specific workflow material.
What to review
Instruction structure, configuration organization and verification conventions. These repositories document developer workflows; the showcase alone is not evidence of a hosted autonomous agent service.
Explore showcase · Review ecosystem · OpenCode workflows
Backend experiments · Python / FastAPI / pytest
AI SDK adapters — mission routing with explicit contracts
Problem
Represent a technical mission consistently across API and CLI entry points while exploring different AI SDKs.
Implementation
Shared skill-aware routing, normalized mission inputs and outputs, FastAPI health/run endpoints, CLI runners and contract tests. The LangChain adapter renders a ChatPromptTemplate; the Anthropic adapter initializes its SDK client lazily.
Current limits
Connecting routed missions to Claude Messages API calls remains on the Anthropic roadmap. Provider execution and tracing remain incomplete in the LangChain adapter. These are source-available experiments, with no public interactive demo promised.
Anthropic source · LangChain source
Delivery infrastructure · TypeScript / Cloudflare / Vercel
Portfolio Time Router — one identity, five specialist views
Problem
Present different engineering disciplines through one portfolio network while keeping a stable professional biography and dedicated URLs.
Implementation
Time-based portfolio selection, a sticky visitor cookie, a manual portfolio selector, origin fallback and dedicated discovery endpoints. A stable biography and canonical profile connect the specialist sites.
What to review
Routing behavior, error handling, tests and consistency between the Worker and Vercel implementations. The project demonstrates deployment and routing decisions rather than a measured scale claim.
Visit portfolio network · Review router source