About/hellpuffyt
Nepal
PrabeshSharma
01Full-stack & backend
02Systems software
03Developer tools
What I build
I build business software for companies in Nepal through Data Doodles Tech. In the open I write systems software and tools: a compiler, a VM, a temporal database, an HTTP server and a set of small CLIs.
- Full-stack engineering
- Cloud infrastructure
- Cybersecurity
- AI / ML
- SaaS
- Automation
- Developer tooling
- Travel technology
- Business software
What I care about technically
- 01
Explain, don't just report.
A finding is only useful if it names the consequence. My tools say which production failure a manifest invites, which users an accessibility defect locks out, and what a packet flow means, in plain English.
Evidence
- 02
Understand the whole stack.
A compiler with its own x86-64 backend, a register VM, a temporal SQL database and an HTTP/1.1 server built on RFC 9112, each with no dependencies or only the standard library, so nothing is hidden behind a framework.
Evidence
- 03
Determinism is a feature.
Systems that can be replayed can be trusted. Gas-metered execution, record/replay of every side effect, and rewindable simulation make behaviour inspectable instead of mysterious.
Evidence
- 04
Software should refuse to lie.
Status derived from real checks, never hand-edited. Extraction that reports what it could not parse. Inference that claims only what the evidence supports. Honest output is a design constraint, not a disclaimer.
Evidence
Engineering philosophy
Build the thing, then explain it. Most of my repositories exist because I wanted to understand a layer properly: how a compiler lowers to machine code, how a database keeps history, how an HTTP server actually parses a request. Writing one is the fastest way I know to understand a system.
Small, sharp tools over sprawling ones. A linter that names the production failure. A detector that reports its own limits. A status page that can't be hand-edited into lying. Narrow scope makes honest output possible.
The interface is part of the engineering. Whether it's a terminal report, a dashboard or this site, the way a system explains itself decides whether anyone trusts it.