Azhar Alhasan

AZHAR ALHASAN

Staff Machine Learning Engineer · Malmö, Sweden · Arabic | English | Swedish
I ship production GenAI, RAG, and AI systems that handle millions of interactions, improve resolution rates, and lower costs. Recent wins: IKEA's 63-country bot (+35% resolution, -20% support cost), tenant-contained LLM chat with privacy/governance deployed in under 24h for sub-$1k/mo, and prompt-based AI playlists driving +25% engagement and -40% curation effort.

Experience · Products · Contact

Signature outcomes: IKEA Billie at 20M+ monthly conversations with +35% resolution and -20% support cost across 63+ markets. Axis Object Analytics shipped portfolio-wide on-device detection/tracking at no extra cost. Spotify's AI Playlist delivered +25% engagement with -40% curation effort.

What I do: production GenAI across chat and voice with eval loops, latency and cost controls, and observability; tenant-first LLM and RAG with privacy and governance guardrails; AI features that ship from playlists to support automation and on-device intelligence.

How I work: discover the right outcomes with privacy and infra alignment, prove with eval harnesses targeting quality, latency, and cost, then ship and learn through autoscaling, guardrails, and steady iteration.

Experience

IKEA logo

IKEA

Oct 2024–Present · Malmö

Staff Machine Learning Engineer · Retail · Dec 2025–Present

Team size 12 · Reporting to Retail Data & Machine Learning Manager

  • Led Billie, a 24/7 AI voice/chat/email assistant handling 20M+ monthly conversations.
  • Scaled coverage to 63+ markets with measurable resolution and cost impact.

Senior Machine Learning Engineer · Retail · Oct 2024–Dec 2025

  • Built eval pipelines that lifted resolution rates 35% and cut support costs 20%.
  • Ensured zero downtime with caching, autoscaling, and end-to-end observability.
Capgemini logo

Capgemini

2022–2024 · Malmö

Senior Machine Learning Engineer · Consulting

  • Built Hitachi Energy Enterprise Chat: tenant-contained LLM chat on Azure with privacy/governance.
  • Shipped guardrails, eval harnesses, and extension hooks for tailored prompts and personas.
  • Delivered Spotify AI Playlist on Premium mobile; +25% engagement, -40% curation effort.
Axis Communications logo

Axis Communications

2019–2022 · Lund

Machine Learning Engineer · Video Surveillance

  • Shipped AXIS Object Analytics across the camera portfolio at no extra cost.
  • Optimized CNNs for ARTPEC SoCs; scaled real-time models to 76% of cameras.

Core Skills

GenAI & RAG

Agentic RAG, prompt engineering, LangChain, LangSmith, semantic search, and Elasticsearch (vector search).

Serving & LLM stack

Vertex AI model providers, LiteLLM routing, Kubernetes microservices, Docker, and Pub/Sub messaging.

MLOps & Infra

GitHub Actions CI/CD, IaC with Terraform, Redis caching, and production eval/guardrail workflows.

Data & services

Postgres for production data, BigQuery Lakehouse, and ClickHouse analytics.

Backend

Python, SQL, and TypeScript with API design, evaluation harnesses, and production guardrails.

Monitoring & observability

Grafana dashboards, Langfuse traces, OpenTelemetry instrumentation, and operational model monitoring.

Products

IKEA logo

IKEA Billie — Support Automation

2024–Present

24/7 customer chat and voice bot across 63+ markets, handling order status, store hours, delivery rescheduling, and issue reporting.

Highlights

  • Feedback-driven evaluation pipelines over 20M+ monthly conversations; +35% resolution and -20% support cost.
  • Latency-aware routing, autoscaling, and caching to stabilize peak loads; observability via Langfuse + OpenTelemetry into Grafana.

Stack

Microservices (Python/Node.js)Vertex AI model providers + LiteLLMLangChain + LangSmithLangfuse + OpenTelemetryRedis caching + autoscaling + TerraformGrafana dashboardsElasticsearch (vector search) + RAG guardrailsPostgres + BigQuery Lakehouse + ClickHouse analyticsCloud: GCP + Azure + AWSKubernetes + Docker + Pub/Sub messagingCI/CD: GitHub Actions
Hitachi Energy logo

Enterprise Chat — Hitachi Energy

2023–2024

Hitachi Solutions Enterprise Chat: LLM chat in your Azure tenant with privacy, governance, and user management—deployable in under a day for sub-$1k/mo.

Highlights

  • Secure SaaS in your Azure tenant with ISO-aligned controls and data privacy.
  • Fast start: up and running in <24 hours; low entry cost (<$1k/mo) with enterprise options.

Stack

Self-hosted LLMElasticsearch (vector search) + PostgresBigQuery Lakehouse + ClickHouse analyticsLangChain + LangSmithVertex AI model providers + LiteLLMKafka + RedisKubernetes (self-hosted) + DockerGuardrails + human-in-loopObservability: OpenTelemetry + Grafana + alertingSecrets/KMS + network isolationTerraform / HelmCloud: AWS + AzureCI/CD: Azure DevOps
Spotify logo

AI Playlist — Spotify

2022–2023

For Premium on mobile: AI Playlist builds a playlist from your typed prompt, curating songs to match your taste.

Highlights

  • Use prompts that mix places, activities, moods, genres, artists, decades, or emojis—combinations perform best.
  • Go to Your Library > Plus > AI Playlist; choose a suggested prompt or type your own, then Add/Save tracks you like.

Stack

Microservices (Python/Scala/Node.js)Kafka + RedisPostgres + BigQuery LakehouseClickHouse analyticsCassandraDocker + KubernetesTerraformPrometheus + GrafanaLLM routing · realtime analyticsEval + retraining loopsCloud: AWS + GCPCI/CD: Backstage
Axis Communications logo

AXIS Object Analytics

2019–2022

On-device AI across the Axis camera portfolio; detection, tracking, and classification shipped at no extra cost.

Highlights

  • Optimized CNNs for ARTPEC SoCs (quantization, memory, throughput).
  • Enabled concurrent real-time models on constrained hardware.

Stack

Edge CNNsQuantizationThroughput optimizationC++/Rust inference pipelineOn-device inference + real-time trackingCloud: Azure + AWSObservability: Prometheus + GrafanaTerraform / HelmCI/CD: Jenkins

Education

MSc, Embedded and Intelligent Systems

Halmstad University · 2017–2019

BSc, Electrical and Electronics Engineering

University of Basrah · 2012–2016

Let's Talk

Reach out to explore future collaborations—whether you need GenAI, RAG, or ML leadership, or want a quick gut-check on where to take your stack next.