Junaid Maqbool

Full-Stack TypeScript Engineer

Lahore, Pakistan

JunaidMaqbool

Full-Stack TypeScript Engineer

Four years on production SaaS: reliability work on systems that had stopped being reliable, AI pipelines at real scale, and small teams under real deadlines. Every number in this file has a document, a dashboard, or a repository behind it.

1 JOB · 3 PROJECTS · 3 FRAMEWORKS

junaid.maqbool046@gmail.com LinkedInLahore, Pakistan

01/09

JunaidMaqbool

Full-Stack TypeScript Engineer

1 JOB · 3 PROJECTS · 3 FRAMEWORKS

junaid.maqbool046@gmail.com LinkedInLahore, Pakistan

Software Engineer at Lyftrondata

JobJul 2022 — Oct 2024 · Remote

First job out of university. Two years on the parts of the product that weren't allowed to break.

INSIGHT

Qcluster wasn't slow. It was losing work silently, which is worse — nobody knows to retry what they never saw fail.

Dead tasks

Pipeline efficiency

Recognition

Star Player 2023

Data EngineeringApache AirflowPython / DjangoAzureCode Review

Hover either document to read it full size.

The Platform Was Failing Eight Requests Out of Ten

ProjectCase #01

100,000 users a month, an extraction pipeline the platform had started blocking, and a vendor fix that didn't hold.

INSIGHT

The 60% ceiling was the tell. It meant the proxies were fine and the rotation was the bug.

Error rate

Infra cost

Monthly users

100,000+

Reliability EngineeringInfrastructurePythonOrganic Growth

Built withPythonFastAPICeleryCelery Beat & FlowerInngestNext.jsDrizzle ORMStripe

Leading an Interview Platform From V1 to 6,000 Users

ProjectCase #02

Started as a two-person side project. Ended as a three-person team, four models, and three full rebuilds.

INSIGHT

One model for everything is a bill, not an architecture. Which model does which task is a cost decision as much as a quality one.

Active users

Paid subscribers

Peak monthly visitors

75,000

AI/LLM IntegrationTeam LeadershipPrompt EngineeringGrowth Experimentation

Built withClaudeWhisper on GroqGemini (Nano Banana)GPTshadcn/uiSupabaseWasabi S3Next.jsStripe

An AI Content Engine That Marks Its Own Homework

FrameworkCase #03

A pipeline that researches a topic, writes it, grades itself, and refuses to ship until it passes.

INSIGHT

Put the expensive model on the quality gate and the cheap one on generation. Judging is where accuracy pays; writing at volume is where cost does.

Pages live

Pages targeted

Countries covered

119

AI PipelinesPythonProgrammatic SEOCost-Aware Architecture

How it worksTwo-model quality gate (9/10 threshold)Anti-overfitting sample rotationCost-tiered model routingSelf-healing internal linkingState-aware pipeline CLI

Making a Junior Engineer's Next Project Impossible to Get Lost In

FrameworkCase #04

He'd just failed a task under close supervision. I fixed the spec instead of watching him harder.

INSIGHT

Junior engineers don't fail at typing. They fail at the uncomfortable part of research, where you have to ask yourself the right next question with nothing to go on.

Build time

Meetings needed

Deep code review

Not required

Technical LeadershipMentorshipSystems DesignAI-Assisted Workflows

How it worksCompetitive-teardown skill (paired human/AI)Schema-first specificationPhased, dependency-gated deliveryNamed review triggers

Governing an AI Coding Agent Like a Senior Engineer Would

FrameworkCase #05

Nobody asked for this one. I built it because the same friction kept showing up on every project.

INSIGHT

An agent that can't say "I don't understand this yet" will guess instead. The confidence score exists to make guessing visible before it becomes code.

Origin

Approval gate

Confidence bar

≥9.5/10

AI EngineeringDeveloper ToolingProcess DesignSelf-Initiated

How it works7-phase execution pipelineConfidence threshold (≥9.5/10)Risk-weighted test proposalsDependency-graph code reviewMandatory automated QC gateFigma-to-code design tokens

Testing Every Option Before Betting the Product on One

ProjectCase #06

I evaluated ten-plus vendors against the detection tools users would actually face, then led the six-week build without writing the code.

INSIGHT

A vendor that looks good in a demo and a vendor that survives the tool your user is graded by are two different vendors. Only one of those tests is worth running.

Signed-up users

APIs evaluated

Time to ship

~6 weeks

Vendor EvaluationAPI ArchitectureProject LeadershipEvidence-Based Decisions

Built withNeon PostgresNext.jsDrizzle ORMBetter AuthStripe

CASE CLOSED

LET'S TALK

Thanks for reading this far. That's more attention than most people give a resume, and I appreciate it. If anything in here was worth a conversation, I'd be glad to hear from you.

junaid.maqbool046@gmail.com LinkedIn

Education

B.S. Computer Science
PMAS Arid Agriculture University
2018 — 2022 · CGPA 3.2/4.0