Anas Qumhiyeh

Anas Tarek Qumhiyeh

AI Engineer, Deriv

Bandar Sunway, Selangor, Malaysia. Malaysian citizen. Open to remote. Open to relocating to Singapore or Australia.

Summary

Full-stack engineer building production services in Go, TypeScript and Python on PostgreSQL, Redis and Kubernetes, with digital-commerce experience in payment-gateway integration and merchant onboarding.

Experience

Jun 2026 - now

Deriv

AI Engineer · Full-time

Go, FastAPI, Next.js, Kafka, Terraform, AWS EKS/EC2

  • Shipped a full-stack customer-support system that proactively resolves user issues.
  • Built a Go microservice that unifies data from 4-6 internal services into a single response for the support system, now running in staging.
  • Built and operate Kafka- and API-based ingestion pipelines for reliable downstream analysis.
  • Built evaluation and debugging harnesses that reproduce failures and capture environment signals.
  • Updated the team's Terraform configurations while scaling existing infrastructure and provisioning new services.
  • ~80% lower CX costProduction
  • CSAT 5/10 → 8/10Production
  • ~20,000 complex events/dayProduction
  • ~50% faster time-to-fixProduction
  • ~10-30% lower AWS costProduction
  • 4-6 services → one responseCapability

Jan - Jun 2026

Skribble Lab

Software Engineer · Full-time

Laravel, PostgreSQL, Redis, Docker

  • Owned the Xendit payment-gateway integration end to end, from requirements review through production support.
  • Designed and shipped a merchant onboarding module for product and payment setup, and onboarded the primary client, Chung Ling Private High School.
  • Hunted integration breakage and production defects before customers found them.
  • 5+ merchants on XenditProduction
  • 3,000+ students and parentsProduction
  • Weekly production errors 16 → 3Production

Nov 2025 - Feb 2026

Monash University

Full-Stack AI Engineer · Contract

FastAPI, Neo4j, Next.js, LangGraph

  • Built a FastAPI and Neo4j graph service for university regulations with a self-correcting Text-to-Cypher loop.
  • Automated prerequisite and credit-transfer rule resolution with embeddings plus multi-hop graph queries.
  • Profiled and simplified the reasoning pipeline into a graph-logic SLM that surfaced contradictory policy data for administrators.
  • +45% retrieval vs vector-onlyControlled evaluation
  • −50% inference latencyControlled evaluation

Feb - Dec 2025

Western Digital

Full-stack Engineer · Contract

Next.js, ASP.NET, PostgreSQL, Docker

  • Built the lab-operations dashboard used by 50+ lab staff, with WebSocket station-status and model-inference updates under 100 ms of UI-visible latency.
  • −40% manual oversightControlled evaluation

Jul - Dec 2025

Setel

Software Engineer Intern · Internship

Docker, Kubernetes, React, Node.js, MongoDB, NestJS

  • Authorization and capture of stored payment methods on the payment engine, documented so a new developer could follow the path without a walkthrough.
  • 92.5% unit-test coverageControlled evaluation
  • −40% production defectsProduction

Nov 2024 - Feb 2025

Petronas

Project Engineer Intern · Internship

MATLAB, Python, MathCAD

  • Replaced MATLAB-dependent back-end calculation and reporting functions with Python packages, removing paid runtime dependencies.

Western Digital (Feb-Dec 2025) overlapped with the Setel internship (Jul-Dec 2025) and final-year study. The Monash contract (Nov 2025-Feb 2026) overlapped with the start of Skribble Lab (Jan 2026). I graduated in April 2026 and joined Deriv in June.

Projects

Jul 2026

FaultLine

Regression and leak-evidence CLI · Solo

Go, Docker, GitHub Actions, Sigstore

Replays a frozen, human-reviewed check across Git history in Docker sandboxes to find the first reliable PASS→FAIL commit.

  • First PASS→FAIL commit in 1-5 minControlled evaluation
  • 1-3 h saved per investigationControlled evaluation

Mar 2026

Gemini Teleportal

Voice-controlled desktop AI agent · Built with Kai

Next.js, TypeScript, Python, Gemini Live API, WebRTC, Firebase, Cloud Run

Streams phone audio and screen video to the Gemini Live API, which drives the Windows desktop by exact UI element ids.

  • 90%+ of demo desktop tasks, unaidedControlled evaluation
  • ~2-5 s from voice to actionControlled evaluation

Mar 2026

CircuitMindAI

PCB inspection · Solo

Next.js, Amazon Bedrock, Express, OpenSearch Serverless, GitHub Actions, ECS Fargate

Images in, voice-guided inspection steps out, with cached results for network loss.

  • Vision and voice PCB inspectionCapability

Apr 2026

RexCheck

DeFi pool health scoring · Built with Chang Kai Zhe

Python, asyncio, Redis, Rails 8, Hotwire, PostgreSQL, XGBoost, MCP

New DeFi pools in, a health score and a trade recommendation out, for a live dashboard and for AI agents.

  • Pool health scoring over MCPCapability

Apr 2026

Veridian

MLOps tradeoff engine · Solo

Python, Terraform, BigQuery, GCP Vertex AI, GitLab Duo, MCP

Intercepts Terraform and Kubernetes and recommends a lower-carbon compute configuration before provisioning or a demand-driven scale event.

  • −15% cloud emissionsControlled evaluation
  • 99.9% observed uptimeControlled evaluation

Oct 2025

Multi-Agent GraphRAG

Policy-corpus retrieval · Solo

LangGraph, Neo4j, Vector DB, Knowledge Graphs, RAG

LangGraph over an independent university handbook and policy archive, with a vector fallback.

  • +35% retrieval vs vector-onlyControlled evaluation

May 2025

MirrorFi

Solana vault strategy platform · Hackathon team of 6

Next.js, ShadCN, MongoDB, Solana, Node.js

A no-code builder for sharing and executing Solana yield strategies across Drift, Jupiter and Meteora.

  • Grand Prize, Solana Megahack 2025Award

Jul 2025

Financial Risk Predictor

ML risk assessment · Solo

TensorFlow, XGBoost, LightGBM, BentoML, SHAP, Kafka

Daily credit-risk scores that have to be interpreted, served, and retrained.

  • 0.87 AUC-ROCControlled evaluation

May 2025

Distributed Lead Scorer

Large-scale data mining pipeline · Solo

PySpark, PyTorch DDP, Deep Interest Network

Scores conversion on a hundred million events a day, with checkpoints so a failed hour can resume.

  • 100M-event capacity benchmarkCapacity benchmark

Jul 2025

SLM Distillation Engine

Knowledge distillation and fine-tuning · Lab experiment

PyTorch, QLoRA, DeepSpeed, FlashAttention

A 70B teacher distilled into a 3B student with QLoRA, DeepSpeed and FlashAttention.

  • 70B → 3B studentControlled evaluation

+45% is the Monash contract on university regulations. +35% is the independent GraphRAG project on a separate university handbook and policy archive. Same comparison, vector-only, different corpus. The halved inference latency elsewhere on this site belongs to the Monash contract's graph-logic SLM, not to this project.

Education

Apr 2026

Monash University

Bachelor of Engineering (Honours), Software Engineering

Minor in Artificial Intelligence. WAM 81.8, CGPA 3.78. Bandar Sunway, Selangor, Malaysia.

Awards

  • First Class Honours
  • Best Graduate Award
  • Dean's List
  • Grand Prize, Solana Megahack 2025 (MirrorFi, team of 6, out of 150+ teams)

Skills

Languages
Go, TypeScript, Python, JavaScript, SQL, PHP, C#, HTML/CSS
Frontend
React, Next.js, Tailwind CSS, ShadCN, jQuery
Backend and APIs
Node.js, NestJS, FastAPI, Laravel, ASP.NET, REST APIs, WebSockets, WebRTC, Kafka, Microservices
Data
PostgreSQL, Redis, MySQL, MongoDB, Neo4j, Elasticsearch
Cloud and DevOps
Kubernetes (AWS EKS), Docker, AWS (EC2, EKS), Google Cloud (Cloud Run, Cloud Build, Firestore), Terraform, ArgoCD, GitHub Actions, GitLab CI/CD, Grafana, Loki
AI/ML
LangGraph, PyTorch, TensorFlow, pandas, NumPy

AI Engineer at Deriv since June 2026. Open to conversations about mid-level software engineering roles.