Ujjansh Sundram / Portfolio
SELECTED WORK / 01

ArchRivet

A second look at a backend design. This experimental MCP tool checks the facts an assistant supplies, flags missing details, and returns findings with evidence.

MOTION STUDY / LLM + MCP
01 / 03

The LLM drafts

Your AI assistant drafts a backend design and supplies its explicitly stated facts.

Illustration of the project workflow.

  1. The LLM drafts

    Your AI assistant drafts a backend design and supplies its explicitly stated facts.

  2. ArchRivet checks

    The local MCP plugin checks the supplied facts, returning scoped findings, evidence and missing details.

  3. Clarify and revisit

    Your assistant explains the findings, asks for a missing detail and can submit a revised design. This is not a deployment test.

On the workbenchTry the retry check

In the example, the worker can receive the same message twice. ArchRivet checks whether the store enforces the message ID and commits it alongside the write. Read the source

PythonFastAPIMCPPydantic

The problem

A convincing architecture description can still leave retries, durability, and overload behavior unspecified.

The approach

Local rule checks turn supplied design facts into scoped findings, questions, and workload calculations. Revised facts can be reviewed again.

The result

An experimental plugin with structured evidence for the next design discussion. It does not inspect a deployed system or certify its safety.

Explore repository