Senior Software Engineer

Mayur Darji

Portrait of Mayur Darji

Nine years in software development.

My career includes cybersecurity DoD contracting, running an IoT startup as a technical founder, and building cloud software at Microsoft.

I currently work in security software operations, build AI agentic tooling, and explore other domains on the side.

Washington, D.C. Remote

Get in Touch by email

Developer tools

  • An issue tracker for your terminal and your coding agents. All of your agents across devices can track markdown tickets and boards.

    Inside the build: Atlas Tasker

    Tickets are Markdown, history is an append-only event log, and SQLite is a rebuildable index. The work stays with the repo. Leases and review gates make ownership explicit when several agents work at once. Agents use CLI help or MCP tools to find work, claim tickets, and update status with machine-readable results.

  • My way of executing loops in multi-agent workflows. Coordinates planning, implementation, and review through durable queues, saved artifacts, and explicit approval gates.

    Inside the build: Arthur Loop

    Loop state, queues, and saved artifacts live in Markdown and JSONL files. Planner, implementer, reviewer, and QA roles can use different coding agents, with CLI and MCP tools advancing the same queue. Implementation gates and project-scoped human decisions make handoffs explicit.

  • Tools that go in front of AI workload processing that I commonly use in forward-deployed engineering. Static verifications, safeguards, and processing.

    Inside the build: Field Kit

    Eight Python plugins share a local CLI and a browser hub bound to localhost. They profile files, replace detected PII, generate seeded demo data, compare datasets, and check drafts and AI workload specs. JSON reports support agent workflows, while Netwatch can observe and gate HTTP connections through a local proxy.

  • Give the agent a stand-in key. Keep the real one encrypted. A local API key broker for apps and coding agents.

    Inside the build: Slifer Key Locker

    A loopback-only broker checks token scope and policy before decrypting a provider key and attaching it to an outbound request. Apps hold Locker tokens instead of raw provider keys in proxy mode. Private pre-release; there is no public installable release, and it is not ready for production secrets.

  • Know when your agent needs you. Bring your terminal and AI-agent sessions into the browser, with tmux keeping work running when you close the tab.

    Inside the build: Helioterm

    A local daemon connects the browser to durable tmux sessions, so closing a tab leaves the underlying work running. One device controls a session while others can observe. Private remote access uses Tailscale Serve, with application login still required. The project is in pre-release.

Research tools

  • A research experiment where I'm trying to teach agents different types of "taste" and "style" and hopefully this can spark some more creative output.

    Inside the build: NanoTaste

    Markdown taste files guide a local Python CLI that compares candidate outputs and records its choices in JSONL. Coding agents can call the CLI, and opted-in session history can propose taste rules for review. The current critic uses inspectable lexical rules; it remains an experiment in output selection.

Mayur Darji — Senior Software Engineer