I build software that connects systems-level engineering with practical developer workflows: desktop AI agents, autonomous coding runtimes, Linux tooling, infrastructure, and applied ML research.
My work focuses on making complex software systems more useful, inspectable, and dependable. I care about clear architecture, explicit safety boundaries, reproducible verification, and honest documentation of what is implemented versus planned.
|
Local-first desktop AI agent A Tauri, Rust, and React desktop assistant spanning everyday computer tasks and developer workflows. The project explores visual screen understanding, browser automation, reversible file operations, document assistance, build diagnostics, and UI auditing. Stack: Repository · CI |
Autonomous software engineering runtime A Rust-native terminal coding agent built around a key boundary: the model proposes; the runtime decides. Its design emphasizes governed tool execution, recoverable workflows, checkpoints, and evidence-based verification. Stack: Documentation · Releases · CI |
|
Linux systems and desktop experience An Arch-based Linux ecosystem focused on system tooling, desktop workflows, and balancing performance with elegance. Focus: |
Terminal-native AI code review A developer tool combining deterministic repository analysis with optional model-backed review, interactive terminal workflows, and machine-readable output for automation. Stack: Repository · Releases · CI |
- OpenMail — self-hosted organizational webmail built to connect with existing mail infrastructure.
PHP·Laravel·Livewire - SLMGen — tooling for dataset analysis and small-language-model experimentation and fine-tuning workflows.
Python·FastAPI·TypeScript - Runstead — a hosting-business management platform with a browser-first interactive simulation. Some infrastructure and payment flows are demonstrations, not live operations.
Next.js·TypeScript - Tonmoy Infrastructure & Vision — the broader portfolio spanning software, developer infrastructure, and research.
Project maturity varies. Please use each repository's current README, release history, and CI results as the source of truth for supported features and release readiness.
- Architecture first: keep boundaries and responsibilities explicit.
- Evidence over claims: prefer reproducible tests and observable behavior over vanity metrics.
- Safety by design: validate untrusted input and put meaningful controls around consequential actions.
- Recovery is a feature: design for cancellation, checkpoints, rollback, and useful diagnostics.
- Honest status: distinguish implemented behavior, demonstrations, experiments, and roadmap items.
| Area | Tools and topics |
|---|---|
| Systems engineering | Rust, Go, Linux, shell tooling, process and runtime design |
| Developer experience | TypeScript, React, Next.js, terminal interfaces, code intelligence |
| Desktop & automation | Tauri, browser automation, computer-use workflows, MCP |
| Infrastructure | Git, CI/CD, databases, deployment automation, operational tooling |
| Applied ML | Small language models, dataset pipelines, evaluation and reproducible experiments |
I prefer links to the source of evidence instead of static claims that can go stale.
| Project | Automation | Releases / source |
|---|---|---|
| Kyro | GitHub Actions | Repository |
| M31A | GitHub Actions | Releases |
| Octate | GitHub Actions | Releases |
| OpenMail | GitHub Actions | Releases |
| SLMGen | GitHub Actions | Repository |
| Runstead | GitHub Actions | Repository |
These links expose current project activity; they do not imply every workflow is passing.
I value focused issues, reproducible bug reports, reviewable pull requests, and documentation that describes actual behavior.
For bugs or feature requests, please open an issue in the relevant project's repository so the context and reproduction steps stay with the code.
Maintained by Eshan Roy · Built in public, with an emphasis on verifiable behavior, thoughtful systems design, and honest engineering.