Skip to content
View rakshithmuda22's full-sized avatar
1 follower · 0 following

Achievements

Achievement: Quickdraw

Achievements

Achievement: Quickdraw

Highlights

  • Pro

Block or report rakshithmuda22

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
rakshithmuda22/README.md

Hi, I'm Sai Rakshith Muda

I build AI agents, and the tools that keep them honest.

AI engineer and founder of VitalMaa. B.S. Computer Science, Penn State '26. Open to AI engineering roles. rakshithmuda22.github.io · LinkedIn · rakshithmuda@gmail.com

What I'm building

Project What it does Proof
Belay Open-source firewall for AI agents. Checks every tool call from Claude Code or any MCP client against a policy before it runs, including order-aware rules like "tests must pass before git push". pip install agentbelay 114 tests · in a prompt-injection test with a local model, 8/8 runs leaked a planted fake key without it, 0/8 with it
agentspec Deterministic, pytest-style contracts for LLM agent tool calls: ordering, budgets, forbidden tools, argument rules, approval gates. pip install llm-agentspec 173 tests · ~2.7 µs per live check
agentprobe GitHub Action that enforces architecture rules on every PR and catches behavior changes with differential property tests. No LLM. 149 tests · v1.0.0 · blocked a real PR
stacktrace-time-machine Sentry error → GitHub GraphQL blame on the crash line at the deployed release SHA → the PR that last wrote it → Slack. Named the bug-introducing PR in 80% of 223 real crashes (12 OSS projects) · 115 tests
informalid A Claude tool-use interview that turns a cash-only worker's answers into a validated financial profile and screens it against 10 Indian government schemes. 127 tests · scheme rules sourced from PIB
codesage Four concurrent LLM reviewers that post inline PR comments on the exact changed lines. 136 tests

Also: Documind (RAG over PDFs with a measured retrieval eval) and Job-Assistant (live demo).

How I work

I design the system, build it with AI coding agents, and verify it like I don't trust it. Every repo has a test suite in CI, and every number in a README comes from a command you can run yourself.

VitalMaa is my startup: an AI health companion for Indian families that speaks English, Hindi and Telugu (Flutter, Supabase, Sarvam AI). Private, in development; happy to walk through it.

Pinned Loading

  1. belay belay Public

    Open-source firewall for AI agents: checks every Claude Code or MCP tool call against a policy before it runs, with an audit trail and CI replay.

    Python

  2. agentspec agentspec Public

    Deterministic, pytest-style contracts for LLM agent tool calls: ordering, budgets, forbidden tools, argument rules, approval gates, live enforcement. Zero dependencies.

    Python

  3. agentprobe agentprobe Public

    Deterministic PR governance GitHub Action: blocks architecture-boundary violations, flags naming/forbidden patterns, and catches behavior changes with differential property tests. LangGraph + Tree-…

    Python

  4. informalid informalid Public

    Seven-question interview that turns a cash-only worker's answers into a validated financial profile (Claude tool use + Pydantic) and screens it against Indian government schemes.

    Python

  5. stacktrace-time-machine stacktrace-time-machine Public

    Sentry webhook service that blames the crash line at the deployed release SHA via GitHub GraphQL and posts the PR that last wrote it to Slack. 80% top-1 on 223 real crashes.

    Python

  6. codesage codesage Public

    Webhook-driven AI code review: FastAPI + Redis/RQ + Groq. Four concurrent agents (bugs, security, style, tests) post inline comments on the changed lines of a PR.

    Python