01 MUHAMMAD NAUFAL MAHDI · FULL-STACK ENGINEER · AI SYSTEMS

I build products end to end, then verify the result.

currently deep in harness engineering harness engineeringcontext engineeringprompt engineeringagentic loop orchestrationmulti-model routingcontext window budgetingtool call validationsubagent orchestrationeval harness designMCP server integration

Full-stack delivery is the job. AI systems, agent orchestration, and model routing are leverage for moving faster without giving up ownership.

View EQ-Math case study ↘Open CV ↗Email me

2supervised internships1live product demoOPENjunior / associate roles

Muhammad Naufal Mahdi standing at a tea plantation

OUTSIDE / 01OFF SCREEN

Scroll to inspect↓

Next.js✳TypeScript✳Go✳Python✳Medusa.js✳PostgreSQL✳Redis✳Clerk✳Rust✳Claude Code✳OpenCode✳Codex✳Gemini✳9Router✳Hermes Agent✳Pi Dev✳RAG✳Telegram✳Discord✳WhatsApp✳SQLite✳Next.js✳TypeScript✳Go✳Python✳Medusa.js✳PostgreSQL✳Redis✳Clerk✳Rust✳Claude Code✳OpenCode✳Codex✳Gemini✳9Router✳Hermes Agent✳Pi Dev✳RAG✳Telegram✳Discord✳WhatsApp✳SQLite✳

01Full-stack product delivery02AI workflow orchestration03Deployment and verification

02 EXPERIENCE AND EDUCATION

Supervised work came before AI leverage.

Formal Software Engineering training and real delivery provide the foundation for my current workflow.

Industry placementDecember 2020 - January 2021

iDPremium Service and Store

Software Engineering Intern

Delivered a web point-of-sale tool for day-to-day store operations.

Built the web POS around the store's operational workflow.
Helped staff learn and adopt the system after delivery.

Web POSProduct deliveryStaff onboarding

Industry placement16 October 2017 - 14 April 2018

PT Inosoft Trans Sistem

Software Engineering Intern

Completed a supervised industry placement in a professional software environment.

Practiced HTML, PHP, database work, and user-interface implementation.
Worked within team expectations for discipline, collaboration, initiative, and responsibility.

PHPHTMLDatabaseUI

Education2016 - Graduated 13 May 2019

SMK Negeri 1 Surabaya

Software Engineering (RPL)

Completed the three-year vocational Software Engineering program.

Built a foundation in programming, databases, interfaces, and software delivery.
Applied that training through two supervised industry placements.

Software EngineeringVocational trainingIndustry practice

03 FEATURED EVIDENCE

One product you can open and inspect.

EQ-Math is featured because it has a live demo and a concise evidence-backed case study. Source remains private.

Education product engineering

Private sourcePublic proof ready

EQ-Math

A deployed Indonesian math-learning product with role-based flows, strict TypeScript, SQLite, security hardening, CI, and a public demo.

Next.jsTypeScriptSQLiteSECSecurity

View case study ↗Open live demo ↗

Private systems

02 · Paper-trading research automationSpot Signal Lab

A deterministic research lab that stores evidence bundles and outcome labels, produces reports, and searches prior work with SQLite FTS.

Proof package planned

03 · EVM event pipelineBigswap v2

A Go pipeline that scores EVM events, routes RPC requests, backfills outcomes, sends alerts, and records operational telemetry.

Proof package planned

04 LIVE SHOWCASE

Open source you can click and inspect.

Seven deployed projects with public repos. No tokens required - just visit.

S1Susi Air Pilot App

Aviation PWA

Pilot companion PWA with duty calendar, rolling flight-hour limits, document expiry tracking, and a Gemini-powered AI copilot with mini-RAG and function calling.

Nuxt 4TypeScriptGemini AI

Live demo ↗Source code ↗

S2Open Pixel

Web3 pixel RPG

AI-native Web3 pixel quest RPG with guest-first onboarding, no token economy, and optional safe wallet proof.

Next.jsWeb3Pixel art

Live demo ↗Source code ↗

S3EduAdmin Pro

School admin platform

Multi-tenant Indonesian school administration platform with WorkOS AuthKit, PostgreSQL RLS, role-based access, and AI-assisted grading workflows.

Next.jsWorkOSPostgreSQL

Live demo ↗Source code ↗

S4De'Butter

E-commerce full-stack

Full-stack e-commerce for a Sidoarjo bakery: Medusa backend, Next.js storefront, PostgreSQL, Redis, and Tailwind v4.

MedusaNext.jsTailwind v4

Live demo ↗Source code ↗

S5Pulung

Driving school platform

Driving school demo with Clerk RBAC, QRIS mock payment, student dashboard, PDF invoices, and admin exports.

Next.jsClerkQRIS

Live demo ↗Source code ↗

S6EduQuest

Game-like learning

Gemini-powered game-like learning quest prototype with interactive challenges and adaptive difficulty.

Gemini AIGameEducation

Live demo ↗Source code ↗

S7Migrate

Scrollytelling portfolio

Next.js scrollytelling portfolio with React Three Fiber 3D ID card physics, sequence scroll animation, and dual-theme support.

R3FNext.jsPhysics

Live demo ↗Source code ↗

05 HOW I WORK

Orchestrate. Verify. Optimize.

Generated output stays untrusted until engineering evidence agrees.

Orchestrate

I split work into scoped agent roles with ownership boundaries, provider routes, limited permissions, and explicit handoff points.

Roles before prompts

Verify

I treat generated output as untrusted until lint, tests, builds, security checks, diffs, and live behavior agree.

Evidence before claims

Optimize

I route models, reuse cache, compress responses, and keep tool output compact so context remains an engineering budget.

Cost and context telemetry

Observed routing flow

Request→9Router→Model / provider→Cache + telemetry

LIVE TOOLING

The terminals I live in.

Faithful recreations of my actual daily setup - click a window, replay the session, swipe between agents.

* Claude Code

Claude Code v2.1.211

Welcome back Katso!

Fable 5 with high effort · Claude Max ·

[email protected]'s Organization

~/mast3r

Tips for getting started

Ask Claude to create a new app or clone a repository

What's new

Added `--safe-mode` flag (and `CLAUDE_CODE_SAFE_MODE`) to start Claude Code with all customizations disabled

Added `/cd` command to move a session to a new working directory

Added a `disableBundledSkills` setting to hide bundled skills

/release-notes for more

L SessionStart:startup says: # claude-mem statusThis project has no memory yet. The current session will seed it; subsequent sessions will receive auto-injected context.❯ ship the terminal showcase section⏺ Reading the spec and routing implementation to a worker.⏺ Bash(opencode run --auto -m zai-coding-plan/glm-5.2 ...)⎿ worker started (session t-showcase)✻ Brewing... (esc to interrupt)⎿ worker: 3 files written, lint clean✳ Brewed for 3s❯ ● high · /effort

OpenCode

opencode

Ask anything... "What is the tech stack of this project?"

Sisyphus - Ultraworker · GLM-5.2 Z.AI Coding Plan · max

tab agents ctrl+p commands

Codex

→ cv-site git:(main) ✗ headroom wrap codex

Setting up rtk for Codex...

Proxy already running on port 8787

Dashboard: http://127.0.0.1:8787/dashboard

Launching CODEX (API routed through Headroom)...

OPENAI_BASE_URL=http://127.0.0.1:8787/p/cv-site/v1

>_ OpenAI Codex (v0.144.5)

model: gpt-5.6-sol high /model to change

directory: ~/cv-site

Tip: [tui.keymap] in ~/.codex/config.toml lets you rebind supported shortcuts.

› tighten the diff on terminal-showcase.css• Working (1s ... 3s)• Done. Removed 40 redundant lines, kept the scroll-snap intact.› Improve documentation in @filename

gpt-5.6-sol high · ~/cv-site · Context 5% used

UI recreations of tools I use daily. Not affiliated with Anthropic, SST, or OpenAI.

Supporting evidence

Usage sources remain available for audit, but they are not product outcomes and are never added together.

05 WHY HIRE ME

Numbers that survive an interview.

Verified telemetry from real tools. The baseline bar compares a method, not a person.

larger is better

Local AI coding tokens (input + output)Tokscale snapshot, July 2026. Not a product outcome; never summed with other telemetry.
17.1B
Daily AI coding streakTokscale snapshot. Heading toward 60 days; counts local sessions only.
51d
Model requests routed9Router 60-day routing snapshot. Volume only, not a savings claim.
65,908
Systems shipped end to endEQ-Math (deployed demo) + public showcase repos (Pulung, Debutter, EduQuest) + Bigswap v2 + Spot Signal Lab.
5
Distinct model families orchestratedClaude, Codex (OpenAI), Gemini, plus the 9Router routing layer between them.
4
Method baseline - solo workflow without agentic toolingIllustrative comparison of approaches, not a measured coworker. Lower is the point.BASELINE
1x

06 THREE MILESTONES

01 / FOUNDATION

SMK and supervised delivery (2016 - 2021)

Software Engineering training led into two industry placements and a web POS used by store staff.

02 / INDEPENDENT

Products and open source across the stack (2021 - 2026)

I kept building web products, automation, education systems, Web3 pipelines, open-source showcase repos, and deployment workflows.

03 / NOW

AI systems with verification (2026+)

Now I use agent orchestration as leverage while keeping ownership, tests, security, and delivery grounded in engineering practice.

07 ATS CV

Need the compact version?

Full-Stack Engineer · AI Systems · Remote · Surabaya, Indonesia

A clean one-page CV with exact education and internship dates, selected work, skills, and contact details.

Open CV PDF ↗Download CV ↓

08 OPEN CHANNEL

Working on something difficult?

I am open to remote junior or associate full-time roles, paid internships, and clearly scoped collaborations where AI leverage meets grounded engineering.

[email protected] ↗

LinkedIn GitHub Discord Telegram

Outside the terminal: coffee craft · poultry farming · gardening and zero-waste practice.