About

Fadhil

ML engineer

based in: Stockholm, Sweden 🇸🇪
building: Agentic chatbot and personal assistant
learning: Agent loop and harness engineering
reading: The Book You Wish Your Parents Had Read
listening: Tom Misch
brewing: AeroPress

profile.json1 item

{
  "name": "Fadhil Mochammad",
  "location": "Stockholm, Sweden",
  "now": {
    "building": "Agentic chatbot and personal assistant",
    "learning": "Agent loop and harness engineering",
    "reading": "The Book You Wish Your Parents Had Read",
    "listening": "Tom Misch",
    "brewing": "AeroPress"
  }
}

Education

EE → ML

Electrical engineering at ITB in Bandung, then a master’s in machine learning at KTH in Stockholm.

2018 — 2020: M.Sc. Machine LearningKTH Royal Institute of Technology · Sweden
2012 — 2016: B.Sc. Electrical EngineeringInstitut Teknologi Bandung · Indonesia

edu.json2 items

[
  {
    "degree": "M.Sc. Machine Learning",
    "school": "KTH Royal Institute of Technology",
    "country": "Sweden",
    "years": "2018 — 2020"
  },
  {
    "degree": "B.Sc. Electrical Engineering",
    "school": "Institut Teknologi Bandung",
    "country": "Indonesia",
    "years": "2012 — 2016"
  }
]

Experience

8 roles

From a robotics lab in Bandung to machine learning work in Stockholm.

Sep 2025 — Now: Machine Learning EngineerElectrolux · StockholmAgentic internal knowledge assistant: evaluation architecture, retrieval flow, feedback-triage agent. MLOps for a supply-chain optimization system.
Dec 2020 — Sep 2025: Sr. Machine Learning and Experimentation Platform EngineerTraveloka · JakartaTechnical lead for the experimentation platform. Built an ML serving engine and SDK, a Kubeflow training platform, and ranking and bandit systems.
Jun 2020 — Oct 2020: Machine Learning Engineer InternSpotify · StockholmWorked on a multi-armed bandit for notification optimization: policy, reward definition and offline analysis.
Dec 2019 — Jun 2020: Data Scientist InternElectrolux · StockholmCompared Bayesian and frequentist models for sales-demand forecasting, on Databricks with PySpark.
Apr 2019 — Aug 2019: Data Scientist InternScania · StockholmBuilt a recommendation pipeline for maintenance actions on truck diagnostic trouble codes.
Oct 2018 — Aug 2019: Full-Stack Web DeveloperOnTel AB · StockholmBuilt a real-time interactive presentation app for an Ericsson collaboration.
Apr 2018 — Apr 2019: AI EngineerRiset.ai · BandungLed engineers who built a FaceNet-based face-recognition system and took it to production.
Aug 2016 — Dec 2017: Research AssistantITB Advanced Robotics Lab · BandungDesigned data acquisition for coordinated UAV and ground-robot reconnaissance, integrated with ROS.

exp.json8 items

[
  {
    "role": "Machine Learning Engineer",
    "org": "Electrolux",
    "place": "Stockholm",
    "years": "Sep 2025 — Now"
  },
  {
    "role": "Sr. Machine Learning and Experimentation Platform Engineer",
    "org": "Traveloka",
    "place": "Jakarta",
    "years": "Dec 2020 — Sep 2025"
  },
  {
    "role": "Machine Learning Engineer Intern",
    "org": "Spotify",
    "place": "Stockholm",
    "years": "Jun 2020 — Oct 2020"
  },
  {
    "role": "Data Scientist Intern",
    "org": "Electrolux",
    "place": "Stockholm",
    "years": "Dec 2019 — Jun 2020"
  },
  {
    "role": "Data Scientist Intern",
    "org": "Scania",
    "place": "Stockholm",
    "years": "Apr 2019 — Aug 2019"
  },
  {
    "role": "Full-Stack Web Developer",
    "org": "OnTel AB",
    "place": "Stockholm",
    "years": "Oct 2018 — Aug 2019"
  },
  {
    "role": "AI Engineer",
    "org": "Riset.ai",
    "place": "Bandung",
    "years": "Apr 2018 — Apr 2019"
  },
  {
    "role": "Research Assistant",
    "org": "ITB Advanced Robotics Lab",
    "place": "Bandung",
    "years": "Aug 2016 — Dec 2017"
  }
]

Agents & LLMs

5 tools

Mostly from building the agentic knowledge assistant at Electrolux, plus an LLM experimentation pipeline at Traveloka.

Selected work: The agentic assistant serves 6,000 employees a month; I own its evaluation architecture.
Electrolux: Microsoft Agent Framework
RAG · Azure AI Search · Azure OpenAI
Evaluation-driven development
Arize · OpenTelemetry tracing
Traveloka: LLM experimentation pipeline
Related: What I learned tuning retrieval for a multilingual RAG chatbotpost · 2026-10-02

Evaluating an agent honestlypost · 2026-09-10

An agent that triages its own feedbackpost · 2026-08-02

Agentsnote

Evaluation-driven developmentnote

Feedback triagenote

all in this lane

tool-llms.json1 item

{
  "toolkit": "Agents & LLMs",
  "spotlight": "The agentic assistant serves 6,000 employees a month; I own its evaluation architecture.",
  "tools": [
    "Microsoft Agent Framework",
    "RAG · Azure AI Search · Azure OpenAI",
    "Evaluation-driven development",
    "Arize · OpenTelemetry tracing",
    "LLM experimentation pipeline"
  ],
  "related": [
    "What I learned tuning retrieval for a multilingual RAG chatbot",
    "Evaluating an agent honestly",
    "An agent that triages its own feedback",
    "Agents",
    "Evaluation-driven development",
    "Feedback triage"
  ]
}

Data science

9 tools

Most of this comes from leading Traveloka's experimentation platform; the modelling side is from a KTH thesis and internships at Spotify, Electrolux and Scania.

Selected work: Modernization supported 800+ experiments a year across 30+ teams, cut cloud costs 95% year over year, and reduced API latency 90%.
Traveloka: A/B testing platform
Metric semantic layer
Ranking · CatBoost
Traveloka · Spotify: Multi-armed bandits · Thompson sampling
KTH · Electrolux: Forecasting · GAM · LASSO
Electrolux: Bayesian and frequentist models
Traveloka · Electrolux: SQL · BigQuery · PySpark · Databricks
Scania: Recommendation pipeline
KTH: Sentiment analysis · NLP
Related: Deconstructing PlanOut: a coin toss you can repeatpost · 2026-09-30

Bandits in production: Thompson sampling, explained with couponspost · 2026-04-11

One definition of conversionpost · 2026-02-28

A/B testingnote

Metric semantic layernote

Multi-armed banditsnote

all in this lane

tool-experimentation.json1 item

{
  "toolkit": "Data science",
  "spotlight": "Modernization supported 800+ experiments a year across 30+ teams, cut cloud costs 95% year over year, and reduced API latency 90%.",
  "tools": [
    "A/B testing platform",
    "Multi-armed bandits · Thompson sampling",
    "Metric semantic layer",
    "Ranking · CatBoost",
    "Forecasting · GAM · LASSO",
    "Bayesian and frequentist models",
    "SQL · BigQuery · PySpark · Databricks",
    "Recommendation pipeline",
    "Sentiment analysis · NLP"
  ],
  "related": [
    "Deconstructing PlanOut: a coin toss you can repeat",
    "Bandits in production: Thompson sampling, explained with coupons",
    "One definition of conversion",
    "A/B testing",
    "Metric semantic layer",
    "Multi-armed bandits"
  ]
}

ML platform

9 tools

The plumbing behind serving, training and experimentation, on Google Cloud at Traveloka and on Azure at Electrolux.

Selected work: Own deployment, CI/CD, monitoring and operations for a production supply-chain optimizer.
Traveloka · Electrolux: Python
Prometheus · Grafana · Datadog
Traveloka: Go · Redis
Kubernetes · Kubeflow Pipelines
Terraform
Google Cloud
Electrolux: Argo CD · GitHub Actions
Azure · AKS
MLflow experiment tracking
Related: ETags: how an API tells you the data changed under youpost · 2026-10-16

An agent that triages its own feedbackpost · 2026-08-02

A serving SDK data scientists actually usepost · 2025-06-14

Distributed tracesnote

Kubernetesnote

Model servingnote

all in this lane

tool-ml-platform.json1 item

{
  "toolkit": "ML platform",
  "spotlight": "Own deployment, CI/CD, monitoring and operations for a production supply-chain optimizer.",
  "tools": [
    "Python",
    "Go · Redis",
    "Kubernetes · Kubeflow Pipelines",
    "Terraform",
    "Argo CD · GitHub Actions",
    "Prometheus · Grafana · Datadog",
    "Google Cloud",
    "Azure · AKS",
    "MLflow experiment tracking"
  ],
  "related": [
    "ETags: how an API tells you the data changed under you",
    "An agent that triages its own feedback",
    "A serving SDK data scientists actually use",
    "Distributed traces",
    "Kubernetes",
    "Model serving"
  ]
}

System design

4 tools

Services and pipelines built to stay fast and correct under load, from student projects to production at Traveloka.

Selected work: An experiment assignment service sustained 50,000+ requests per second in a 30-minute load test, with P95 under 100 ms.
Traveloka: Go worker pools · high-throughput services
Latency budgets · P95 monitoring
KTH: Kafka · Spark Streaming · Cassandra
OnTel AB: Real-time web app
Related: ETags: how an API tells you the data changed under youpost · 2026-10-16

Deconstructing PlanOut: a coin toss you can repeatpost · 2026-09-30

SSE on Kubernetes: what a service mesh fixes, and what it doesn'tpost · 2025-03-22

Go worker poolsnote

Observabilitynote

P95 latencynote

all in this lane

tool-systems.json1 item

{
  "toolkit": "System design",
  "spotlight": "An experiment assignment service sustained 50,000+ requests per second in a 30-minute load test, with P95 under 100 ms.",
  "tools": [
    "Go worker pools · high-throughput services",
    "Kafka · Spark Streaming · Cassandra",
    "Real-time web app",
    "Latency budgets · P95 monitoring"
  ],
  "related": [
    "ETags: how an API tells you the data changed under you",
    "Deconstructing PlanOut: a coin toss you can repeat",
    "SSE on Kubernetes: what a service mesh fixes, and what it doesn't",
    "Go worker pools",
    "Observability",
    "P95 latency"
  ]
}

Works

Projects · Posts

Projects with public code, the latest posts and research papers.

Projects: Projects with public codemore
Latest posts: ETags: how an API tells you the data changed under you2026-10-16 · 5 min read

What I learned tuning retrieval for a multilingual RAG chatbot2026-10-02 · 12 min read

Deconstructing PlanOut: a coin toss you can repeat2026-09-30 · 11 min read

more
Publications: FPGA implementation of template matching using binary Sum of Absolute DifferenceAditya Rachman Putra · Fadhil Mochammad · Hans HerdianICSET 2016 · IEEE · pp. 13-176th International Conference on System Engineering and TechnologyDOI: 10.1109/icsengt.2016.7849615

Implementation of hazardous chemical gas monitoring system using unmanned aerial vehicle (UAV)Fadhil Mochammad · Aditya Rachman Putra · Bambang Riyanto TrilaksonoICSET 2016 · IEEE · pp. 163-1686th International Conference on System Engineering and TechnologyDOI: 10.1109/icsengt.2016.7849643

work.json1 item

{
  "projects": "Projects with public code",
  "latest_posts": [
    "ETags: how an API tells you the data changed under you",
    "What I learned tuning retrieval for a multilingual RAG chatbot",
    "Deconstructing PlanOut: a coin toss you can repeat"
  ],
  "publications": [
    "FPGA implementation of template matching using binary Sum of Absolute Difference",
    "Implementation of hazardous chemical gas monitoring system using unmanned aerial vehicle (UAV)"
  ]
}

Say hello

contact

Happy to talk about agents, evaluation or experimentation.

email: fadhil.mochammad1095@gmail.com
github: github.com/fadhilmch
linkedin: in/fadhilmch

contact.json1 item

{
  "email": "fadhil.mochammad1095@gmail.com",
  "github": "github.com/fadhilmch",
  "linkedin": "in/fadhilmch"
}