Kazi Musharraf: Projects

Selected engineering work by Kazi Musharraf: LLM applications, MCP developer tools, typed agent workflows and web infrastructure. Review the problem, implementation and source evidence for each project.

AI application

MK QuizFlow

Problem. Turn study material into quizzes and flashcards while keeping a usable path when AI credentials are unavailable.

Engineering approach. Next.js and TypeScript combine local-first storage and deterministic Quick mode with an optional Vercel AI SDK generation route. The API validates inputs and structured outputs, checks per-client rate limits and quotas, and reports provider failures explicitly.

Evidence. Source includes the capability API handler, deterministic generator and unit tests. These are implementation examples; no usage or accuracy benchmark is claimed.

Scope and limitations. AI mode requires configured credentials or a user-provided key. Quick mode uses heuristics and is labeled separately from AI generation.

LLM applicationsStructured outputNext.jsTypeScriptRate limitingLocal-first

Inspect implementation · Repository · Review test source

Developer tools

Vibe TypeScript CLI

Problem. Expose developer workflows through a command-line tool and a Model Context Protocol interface.

Engineering approach. TypeScript primitives cover planning, execution and repository scanning. The MCP server declares tool schemas and dispatches calls to available primitives; a separate approval primitive asks the user before continuing an approved action.

Evidence. The archived repository contains MCP handlers, approval code and a test directory. The case study links to source rather than asserting production adoption.

Scope and limitations. Tool behavior depends on configured primitives and providers. Tests were inspected as source, not executed for this portfolio update; no throughput or task-success benchmark is claimed.

MCPTool callingAI agentsTypeScriptCLIHuman approval

Inspect implementation · Repository · Review test source

Agent workflow example

LangGraph Mission Flow

Problem. Represent a technical mission as a typed graph with explicit routing, planning and verification steps.

Engineering approach. A Python TypedDict carries mission state through a compiled LangGraph StateGraph. The adapter calls shared routing logic and returns plan and verification text, with API and CLI entry points documented in the repository.

Evidence. The archived adapter source shows route, plan and verify nodes. Contract tests are linked for review.

Scope and limitations. This is a scoped orchestration example, not a benchmarked autonomous execution system. Durable checkpoints, recovery benchmarks and approval gates appear in the roadmap rather than the current graph implementation.

LangGraphPythonAgent orchestrationTyped stateFastAPIContract tests

Inspect implementation · Repository · Review test source

Web infrastructure

Portfolio Time Router

Problem. Serve a network of specialist portfolios through a shared domain while retaining dedicated URLs for each discipline.

Engineering approach. The repository documents time-based portfolio routing, with Cloudflare Workers and Vercel deployment options. The specialist subdomains provide direct entry points for AI, full-stack, backend and frontend work.

Evidence. The routing repository and five specialist portfolios provide reviewable examples of the portfolio network.

Scope and limitations. A rotating root page needs careful canonical and indexing decisions. Routing configuration is separate from the content in these specialist portfolio repositories.

TypeScriptCloudflare WorkersVercelWeb infrastructurePortfolio network

Inspect implementation · Repository

More projects and open-source contributions on GitHub.

View GitHub Profile