skip to content
akansha:roy

AkanshaRoy

backend · systems · infrastructure · ai

open to backend, infrastructure & ai roles

i build backend systems and then take them apart to understand them.

logs, queues, storage, the parts of a machine that only matter when something goes wrong. mostly go.

currently learning applied ai and building a small runtime for supervising coding agents.

01now

what is going on

Kept current by hand. It goes stale, which is part of the appeal.

building
Pony — a small Go runtime for supervising several AI coding agents at once
learning
Pony internals: processes, PIDs, sessions, concurrency and coordination
reading
Designing Data-Intensive Applications, slowly and on purpose

as of Oct 2026

02selected workall 7 →

Three things worth reading about

2026

Pony

A lightweight runtime for running and managing several AI coding agents in one process — scheduling, isolation, and a shared view of what each one is doing.

Go · process supervision · PTY · agent tooling

2026

Kofta

A Kafka-inspired distributed log system: segmented on-disk storage, sparse indexes, record batches with CRC validation, consumer groups and a multi-broker coordinator.

Java · log storage · consumer groups · multi-broker

2025

mini-rpc

A small RPC engine in Go over raw TCP — message framing, serialization, method routing, and the client/server boundary, with no framework in the way.

Go · TCP · framing · serialization

03writingeverything →

Notes from the middle of things

19 Apr 2026

i read the kafka paper, then rebuilt it badly

A month of reimplementing Kafka's log layer from the design paper, and the storage engine I had to rewrite three times because I skipped the parts I didn't understand.

10 min · Distributed Systems · Storage · Go · Java

04activity@akansha204 →

A year of work

A simple record of the days I spent building, fixing and learning in public.

active days
149
longest streak
15d
current
8d
lessmore