I take an AI use case from "is this even worth building?" to a deployed, documented, cost-modelled system — solo, end to end.
🌐 Live product: jobiqueue.com ·
GitHub ·
Email ·
Book a call
Dubai / UAE · remote · Top Rated & 100% Job Success on Upwork · GCP ACE certified
Most AI portfolios show you that something was built. This one shows you the judgement around the build — because when one person is responsible for a company's AI, the scarce skill isn't wiring an agent. It's knowing which agents should exist, what they cost to run at scale, and when the right answer is "don't build this."
So every project here is documented through three lenses the work actually depends on:
Interrogate (is the value real and the run-cost viable?) → Build (solo, production-grade) → Watch the meter (model the spend, monitor it, choose the cheapest tier that clears the bar).
Two short reads make that concrete:
- 📊 AI Economics & ROI — the model I use to decide what's worth building, with token/API/multi-agent cost maths for every build.
- 🧭 Use-Case Prioritisation — how I rank a backlog into build / defer / drop with reasoning leadership can stand behind.
- 🔧 Engineering Patterns — what makes these production systems, not prototypes.
| Competency you're evaluating | Where to see it |
|---|---|
| Architecting intelligent systems | AetherFlow — 3 cooperating agents with memory, tools, structured output; 3Distica — agent + tool-calling + sub-workflow orchestration |
| AI automation implementation | All four builds — delivered, solo, end-to-end, in production patterns |
| Accurate ROI calculation & estimation | AI Economics doc + a run-cost table in every README — including the multi-agent cost multiplier most people miss |
| Deployment & scalability | Deltamax — rate-limiting, retries, chunking, defensive parsing; Hamster Garage — pagination, dedup, error capture, cost-at-scale |
Ordered most-sophisticated-AI → most-production-ops. Each is sanitised and importable into n8n.
| # | Build | What it is | Proves | Run cost (modelled) |
|---|---|---|---|---|
| 01 | AetherFlow | 3-agent outbound pipeline: scrape → qualify → compose → reply, each a true agent | Multi-agent orchestration & its economics | ~$60–100/mo |
| 02 | 3Distica | Bilingual (EN/AR) AI sales agent + auto-generated RFQ PDFs | Agentic intake → structured data → generated document | < ~$20/mo |
| 03 | Hamster Garage | Scores 100s of partners against a brief with structured AI, shortlists | AI scoring at scale = RFP/tender evaluation | ~$0.30 per 1,000 |
| 04 | Deltamax | Scheduled time-sync ETL across Hubstaff/Airtable/Sheets | Reliability engineering — and knowing when not to use AI | ~$0 |
Each pattern maps directly onto professional-services / advisory workflows — briefing automation, media monitoring, RFP support, stakeholder reporting — noted at the bottom of every README.
Platform-agnostic by design: these span OpenAI (
gpt-4o-mini), OpenRouter (open models on free/cheap tiers), and no-LLM automation. The wider work below adds Google Gemini Nano (on-device) and Claude (this repo was built with Claude Code). The skill is choosing the right tool per use case — and switching as prices and capabilities move.
When a use case justifies going past workflow automation to a full custom system, I build and deploy those too. Status stated honestly.
- JobiQueue — ✅ live AI job-aggregation platform built solo end-to-end: Next.js + FastAPI + Supabase + GPT-4, aggregating 25+ sources and tens of thousands of jobs, with AI ranking and a published Chrome extension. A deployed AI product with users. → live at jobiqueue.com
- RevOps Autopilot — multi-tenant HubSpot SaaS on GCP (NestJS + Next.js + worker; Cloud Run / Tasks / SQL / Scheduler; Terraform IaC; OAuth, webhook validation, routing engine, audit logs). Built; not currently deployed (runs on paid GCP). →
revops-autopilot-mvp - PR-Pilot — Chrome MV3 extension on Chrome built-in AI / Gemini Nano (Summarizer, Prompt, Translator, Proofreader, Rewriter) with schema-validated output and a Firebase AI Logic fallback. In development. →
pr-pilot - LangMeet.AI — real-time multilingual video calling built solo: live speech-to-text (Whisper) + on-the-fly translation (Google Translate) over WebRTC / LiveKit, on GCP. MVP; the hosted demo is currently offline. Code walkthrough available on request.
Plus 3 years keeping a live European SaaS (hoteliga.com) reliable — Azure, networking, monitoring, incident response. Reliability is a habit, not a buzzword.
n8n · LangChain agents · OpenAI · OpenRouter · Gemini Nano · Claude · Next.js · FastAPI · NestJS · Supabase / Postgres · GCP / Vercel / Fly · Docker · Terraform · Airtable · Google Workspace
Abdullah Javaid
abdullahjavaid.me@gmail.com ·
jobiqueue.com ·
github.com/binary-exe
Every credential, account ID, and secret in this repo has been stripped to a labelled placeholder. The logic and prompts are intact — that's the part worth reading.