Skip to content
View arslanrakh's full-sized avatar

Block or report arslanrakh

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
arslanrakh/README.md

Tech Lead and Platform Architect at Slana.mx, a lending fintech that reached nearly 1M users.

I moved teams off 2-3 week batches onto continuous delivery. Releases later reached 120+ a month, each PR a full system release.

I reworked an inherited distributed monolith of about 40 services for agents and speed: one monorepo, one CI/CD pipeline, the whole codebase as agent context. Each MR got its own reproducible 40+ service stack, replacing shared test environments where only 2 of 6 deployed. The dedicated test clusters (14 nodes, 3 Aurora, 3 MSK) shut down.

Toward the agent frontier

I moved every engineer from building features to engineering agent loops: tuning pipelines, hardening checks and making ADR decisions. A smaller team of professionals outpaced the old-style teams. The goal is for each engineer to ship more at lower cost per change.

I built an AI agent as a virtual employee. It opened and browser-tested MRs in those per-MR stacks. Non-engineers could make changes through it, under code review.

Earlier

At Wallet in Telegram, a 100M+ user platform, I owned P2P transfers on 10-30M Kafka events a day and co-designed anti-fraud with compliance. I persuaded peers to add architecture layers only as needed. Each costs money.

I co-designed a $107M smart grid with General Electric and put four vendor systems on one IEC 61968 model.

Agent-written under my review

  • swarm-forum is protocol R&D for cross-vendor agents, with leases and checkpoints.
  • write-human lints LLM-style prose, with 90% mean detection at 5% false positives on 15 held-out English sets.
  • skills is my set of frontier skills for autonomous software delivery.

Pinned Loading

  1. skills skills Public

    Frontier skills for autonomous agent work: short, sharp principles for smart models, so agents get further with less of you. Built on the best current methods and tested on real agents. For Claude …

    1

  2. write-human write-human Public

    Offline linter that marks LLM-sounding prose. Its statistical model is trained on 2.8B tokens of human text and 11 LLM families, with 0.96 mean AUROC on 15 held-out sets and no model calls at runti…

    Python 13 1

  3. swarm-forum swarm-forum Public

    Self-hosted Rust MCP server where agents from different vendors share work. Fencing tokens, leases and checkpoints let work outlive the worker. Key invariants are checked with TLC, 37 Kani harnesse…

    Rust 1

  4. perfect-recall perfect-recall Public

    Memory for AI agents with no embeddings or vector index. A decision model checks each memory against criteria from the agent. Answer accuracy rose from 48% to 86% with the same caller on 120 LongMe…

    Python 3 1

  5. amcli amcli Public

    ArchiMate models from the command line, no Archi or JVM. One static Rust binary. Every edit is checked against the standard's 62×62 relationship matrix, batches are atomic, and real Archi files rou…

    Rust 2

  6. lid-awake lid-awake Public

    Keeps a MacBook awake with the lid closed, no external display. About 300 lines of Swift with no dependencies or network access. A sudoers rule scoped to two exact pmset commands means one password…

    Swift 1