Skip to main content
Eshan Roy@eshanized
BIOGRAPHY & ENGINEERING STEWARDSHIP

Eshan Roy (eshanized)

Founder & Principal Systems Architect · Tonmoy Infrastructure and Vision

Engineering Background & Focus

I am a systems programmer and open-source architect based in India. Known online as `eshanized`, I specialize in the layer beneath applications: operating systems, compiler-verified autonomous coding runtimes, process sandboxing, and developer infrastructure.

As Founder and Principal Architect at Tonmoy Infrastructure and Vision (TIV), I lead the development of independent, sovereign software systems designed for long-term operational longevity. Our work rejects venture-backed hype and ephemeral trends, focusing instead on structural self-reliance, memory safety, and verifiable systems guarantees.

My engineering work spans three core pillars:

  • Autonomous Runtimes: Creator of M31A (Rust-native autonomous engineering runtime with 11-stage policy gates) and Kyro (local-first desktop AI agent with Tauri v2 and SQLite WAL storage).
  • Operating Systems: Lead Architect of Snigdha OS, an Arch-based Linux distribution ecosystem featuring custom kernel switching tools and automated package mirrors.
  • Systems Tooling & Infrastructure: Author of Polaris (distributed compute framework on crates.io), Veda (adaptive work-stealing parallel runtime in Rust), and SLMGen (dataset fine-tuning infrastructure for Small Language Models).

Verified Identifiers

NameEshan Roy
Public Handleeshanized
Primary LanguagesRust, C, Go, Shell, TypeScript
Direct Emailm.eshanized@gmail.com
[FIRST PRINCIPLES // OPERATIONAL INVARIANTS]

Nine Principles That Govern Software When Hype Fades

These invariants guide every repository, architecture design, and systems experiment I maintain.

01

Architecture before implementation

A codebase without an explicit mental model will inevitably evolve toward maximum entropy. Define layer boundaries, directional dependencies, and invariants before writing a single line of logic.

Applied In: M31A's strict downward L0–L9 hierarchy ensures lower runtime layers never depend on UI or high-level agents.
02

Measure before optimizing

Intuition about performance in systems programming and distributed infrastructure is almost always flawed. Collect profiling data, trace allocations, and inspect cache behavior before introducing complexity.

Applied In: Benchmarking cache-line false sharing in Veda with Criterion before adding 64-byte alignment padding.
03

Explicit failure beats mysterious failure

Silent fallbacks and swallowed errors create operational time bombs. When an assumption is violated, halt, panic, or return an explicit structured error that isolates the failure domain immediately.

Applied In: 11-stage policy evaluation gates in M31A reject unauthorized mutations deterministically rather than degrading silently.
04

Security belongs in the architecture, not as an afterthought

Security cannot be bolted on at the API gateway or wrapped in regex filters. Isolation boundaries, process sandboxing, credential redaction, and least privilege must be primary architectural primitives.

Applied In: OS-level process confinement using Linux cgroups v2, namespaces, and macOS Seatbelt sandboxes built into the core runner.
05

Dependencies should earn their place

Every third-party crate or package is an architectural mortgage that carries supply-chain risk, transitive dependencies, and compilation overhead. If a standard library construct suffices, prefer it.

Applied In: Keeping runtime binaries self-contained with minimal foreign dependencies and deterministic build targets.
06

A system should remain understandable after its author leaves the room

Clever code is a liability. Software must be written for the engineer who maintains it at 3:00 AM during an outage. Clarity of data flow trumps micro-optimizations.

Applied In: State machines with explicit transitions and atomic SQLite transaction event logs that any engineer can inspect with standard tooling.
07

If it can be automated, automate it

Human memory is unreliable for repetitive operational sequences. Package builds, kernel configurations, lint gates, and regression testing must be encoded in reproducible scripts and pipelines.

Applied In: Snigdha OS PKGBUILD automation and headless autonomous execution passes in CI pipelines.
08

If it cannot be observed, it cannot be trusted

Systems running in production must emit telemetry, structured logs, and verifiable state diffs. Black-box operations are unacceptable in critical infrastructure.

Applied In: Structured NDJSON event logs and live telemetry streaming with automated zero-leak credential scrubbing.
09

If it cannot be recovered, it isn't finished

Hard crashes, power loss, and network partitions are normal operating conditions. A robust runtime must detect dirty state on boot and resume deterministically from atomic checkpoints.

Applied In: Two-phase SQLite commits sealing each DAG task with SHA-256 manifests, allowing zero-loss recovery upon unexpected process termination.