Skip to content
View ReaperOAK's full-sized avatar
🏠
Working from home
🏠
Working from home

Block or report ReaperOAK

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
ReaperOAK/README.md
ReaperOAK — Owais Ahmed Khan

Currently

  • Focusing on distributed systems, autonomous AI agents, and backend infrastructure.
  • This cycle — This week: added Claude Code bridge for vibecoding, refactored event handling with stricter types, standardized Vercel and Next.js configs, and updated graphify scripts.
  • Thinking about — Considering event-driven state sync with CRDTs to reduce latency in distributed model serving.

Language mix and coding time

Problem solving progress and deploy status

Featured

Project What it solves Stack
ForgeOS An SDLC engine of orchestrated agents for spec-driven development. TypeScript · Python · JavaScript
csefarewell2k25 TypeScript · Python · HTML
portfolio JavaScript · TypeScript · MDX
vibecoding Python · HTML · Shell
survivorship-free-backtester An honest, survivorship-bias-corrected backtesting engine. Market-agnostic — Indian equities (Nifty 50/100/200/500) wired in out of the box, extensible to US/global via yfinance. Tax-aware, free data, Python + DuckDB. Python · Shell
id-card-maker TypeScript · CSS · Rust

Engineering log

  • New skills were added.
  • Script was updated.
  • Graphify was added.
  • A .claude bridge was added so Claude Code could fully operate the vibecoding system.
  • License was switched from MIT to PolyForm Noncommercial 1.0.0.

Case studies

survivorship-free-backtester — Python · DuckDB · yfinance · NSE bhavcopy

Problem — Retail backtests test the past using today's winning stocks. The bias is invisible and it inflates every result.

Decision — Reconstructed point-in-time Nifty 100/200/500 membership from the Internet Archive, recovered delisted-loser prices, and re-ran every strategy against the honest universe. Lot-level FIFO on top, with real Indian tax (20% STCG / 12.5% LTCG).

Tradeoff — The correction destroyed the headline. Momentum's Sharpe fell from ~1.5 to ~1.0, and after tax the high-turnover strategies barely beat buy-and-hold. Publishing that was the point.

Outcome — Survivorship inflates momentum by +0.4 to +0.9 Sharpe. Ensemble + regime-cash roughly halves drawdown, −38% to ~−18% — insurance, not free alpha. Intraday dip-buying died under honest testing.

Open survivorship-free-backtester ↗

todayeggrates — JavaScript · MDX · PHP

Problem — Daily commodity rates that people actually check, which means the site is worthless the day it goes stale.

Decision — Content as MDX so rates and long-form pages ship through the same pipeline, with SEO treated as a build-time concern rather than a plugin.

Tradeoff — MDX made the build heavier and the content model stricter than a CMS would have. Bought two years of uninterrupted daily updates without an editor UI to maintain.

Open todayeggrates ↗

ForgeOS — TypeScript · Python · PostgreSQL

Problem — Agents given a whole codebase and a vague goal produce plausible output and unreviewable diffs.

Decision — Split the SDLC into stage-specific agents with explicit handoffs and a ticket lifecycle, so each stage has one job and a reviewable artifact.

Tradeoff — Far more orchestration machinery than a single-agent loop, and every stage boundary is a place work can stall. Bought traceability from vision to diff.

Open ForgeOS ↗


Deliberate practice

Track What that means Progress
System design Distributed scheduling, consensus, failure detectors ongoing
NeetCode 150 Pattern coverage, one directory per problem ███████░░░ 101 / 150

The Engine Room

  • Languages · measured — Python 39% · TypeScript 29% · JavaScript 22% · HTML 5% · Shell 2% · CSS 2%
  • Frontend — React · React Native · Next.js · Expo · Tailwind / NativeWind
  • Backend — NestJS · FastAPI · Node.js · Express · Socket.io · BullMQ
  • Generative AI — Claude / GPT / Gemini · Prompt engineering · Multimodal
  • Cloud & DevOps — AWS (EC2/RDS/ElastiCache/S3/Lambda) · Docker · Kubernetes · GitHub Actions
  • Data — PostgreSQL · MySQL · MongoDB · Redis · OpenSearch

Selected numbers

3 yrs shipping production systems · 7M+ search impressions served · ~90% DB load cut by caching · <100ms API responses


How I work

  • Every external dependency is optional. If a provider is down the product degrades, it does not break.
  • A number without a method behind it is decoration. Publish the correction, not just the result.
  • Tests describe behaviour, not implementation. A refactor that breaks the suite means the suite was wrong.
  • Ship the smallest thing that can be measured, then measure it.

Connect

Portfolio • LinkedIn • Email


125R, throttle open · writes poetry · optimizes the economy before attacking

@ReaperOAK's activity is private