Skip to content

Repository files navigation

Systems-lab

A personal learning laboratory for becoming a TypeScript engineer who can build agent-native systems

A deliberate path of hands-on projects in TypeScript + Node.js, ramping from backend fundamentals through the context-engineering and AI-agent infrastructure that today's startups are actually built on: tool-calling loops, memory and summarization, MCP servers, agent orchestration, durable execution, evals, and observability.

Goals

  • Build real backend systems instinct: streaming, concurrency, I/O efficiency, backpressure, observability.
  • Master context engineering: tokenization, token budgets, summarization, hierarchical memory, prompt caching.
  • Become fluent in the agent stack: tool-calling loops, MCP, orchestration, durable execution, sandboxing, and evals.
  • Ship agent-native capstones polished enough to be a remote-hire portfolio.

Completed projects

  • CLI File Analyzer — a robust CLI that analyzes any JSON or CSV file and outputs schema info, row count, null counts, type detection, and basic statistics.

  • Simple Hono API — a basic REST API using Hono with GET routes serving static/mock data, with proper TypeScript and Zod validation.

  • Tokenizer / Lexer — a tokenizer for a slice of JavaScript, built from scratch: scans source character by character into a typed token stream (punctuation, identifiers, keywords, string literals), with whitespace handling and error reporting on malformed input.


Engineering baselines (practices, not deliverables)

Applied across every repo — table stakes, and their absence is what fails take-home reviews:

  • Testing — Vitest for unit tests, Testcontainers for integration tests that spin up real Postgres/Redis. Don't mock what you can run.
  • CI/CD — ESLint + Prettier + a pre-commit hook + GitHub Actions running lint/typecheck/test on every PR.

Tech Stack (Current Phase)

  • Language/runtime: TypeScript (strict) + Node.js
  • Web: Hono
  • Data: Postgres + Prisma, Redis, pgvector, SQLite
  • Queue: BullMQ
  • AI/agents: Anthropic SDK / Vercel AI SDK, MCP, gpt-tokenizer
  • Ops: Docker, OpenTelemetry, Prometheus

Future Plans

Later phases will revisit the performance-critical pieces (tokenizer, sandboxing, storage layer) in Go and C++, and expand into other languages. The focus stays on practical systems thinking, performance, and real-world backend + agent engineering.


Status: In Progress
Started: April 2026


Feel free to explore, star, or fork if you find it useful!

Made with ❤️ for deep learning and systems mastery.

About

No description, website, or topics provided.

Resources

Stars

70 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages