You will get a production-ready ML fraud detection system with 95%+ accuracy
Project details
You will get a production-ready fraud detection system built on a 7-model ensemble — XGBoost, LightGBM, Random Forest, Isolation Forest, Autoencoder, LSTM, and Transformer — achieving AUC 0.9980, Precision 1.0, and zero false positives across 10,000 transactions.
This is not a tutorial or notebook. It is 6,824 lines of production Python across 23 modules, with a FastAPI scoring endpoint, SHAP explainability on every prediction, GPT-4 alert narration, a Streamlit dashboard, Power BI exports, automated reporting, and Azure deployment via Terraform and Kubernetes.
I am an AI and ML engineer based in Oslo, Norway, with two years delivering production systems — AI monitoring for 50+ industrial robots, and a live Azure data platform at 99.5% uptime over ten months.
The system is GDPR and EU AI Act compliant with PII masking, audit logging, and full explainability. AML scenarios include structuring, layering, and smurfing detection. Every decision is traceable and logged.
You will receive full source code, setup instructions, a working API, and a system you can deploy.
This is not a tutorial or notebook. It is 6,824 lines of production Python across 23 modules, with a FastAPI scoring endpoint, SHAP explainability on every prediction, GPT-4 alert narration, a Streamlit dashboard, Power BI exports, automated reporting, and Azure deployment via Terraform and Kubernetes.
I am an AI and ML engineer based in Oslo, Norway, with two years delivering production systems — AI monitoring for 50+ industrial robots, and a live Azure data platform at 99.5% uptime over ten months.
The system is GDPR and EU AI Act compliant with PII masking, audit logging, and full explainability. AML scenarios include structuring, layering, and smurfing detection. Every decision is traceable and logged.
You will receive full source code, setup instructions, a working API, and a system you can deploy.
Machine Learning Tools
Azure Machine Learning, Databricks Platform, Databricks MLflow, Keras, MLflow, OpenCV, pandas, Python, Python Scikit-Learn, PyTorch, SAS, scikit-learn, SciPy, SQL, TensorFlow, XGBoostWhat's included
| Service Tiers |
Starter
$150
|
Standard
$350
|
Advanced
$500
|
|---|---|---|---|
| Delivery Time | 5 days | 8 days | 13 days |
Number of Revisions | 2 | 4 | 6 |
Number of Model Variations | 2 | 4 | 6 |
Number of Scenarios | 3 | 6 | 10 |
Number of Graphs/Charts | 3 | 7 | 11 |
Model Validation/Testing | |||
Model Documentation | - | ||
Data Source Connectivity | - | ||
Source Code |
Optional add-ons
You can add these on the next page.
Real-time streaming pipeline (Kafka)
(+ 3 Days)
+$75
GDPR-compliant Azure deployment
(+ 2 Days)
+$60
Power BI / Looker BI export layer
(+ 2 Days)
+$50Frequently asked questions
About Md Saidul
AI and LLM Engineer - RAG, LLM, Fraud Detection
Oslo, Norway - 1:40 pm local time
For the past two years I've been working professionally in Norway, first at Rolog Solutions as an R&D engineer building an AI-powered monitoring system for 50+ industrial robots, then at Jobswoop as a software engineer running a production data platform on Azure with 99.5% uptime over ten months. Before that, I completed a Master's in Computer Science at UiT — The Arctic University of Norway, where most of my thesis work ended up running in production rather than sitting in a paper.
The kind of work I do: production LLM systems, RAG pipelines, fraud and anomaly detection, agentic AI with tool-calling, EU-compliant cloud infrastructure. Not prototypes — systems that handle real data, real users, and real consequences if something breaks.
Some concrete results: a fraud detection ensemble that reached AUC 0.9980 with Precision 1.0 and Recall 0.96. An agentic RAG platform running on Azure Norway East with GPT-4o tool-calling and real-time streaming, built under Norwegian GDPR. An anomaly detection system that cut false positives by 40% across two million records. A 60% reduction in robot troubleshooting time for a Norwegian robotics company. A 35% reduction in supply chain waste through predictive analytics.
I work across the full stack — LangChain, OpenAI, FastAPI, PostgreSQL, Azure, Terraform, Docker, Kubernetes. I write the models, the APIs, and the infrastructure. I've built to GDPR, EU AI Act, and HIPAA on live systems, so compliance is not something I need to learn on your project.
I'm based in Oslo, which means full working-day overlap with European clients and a reasonable window with the US East Coast. For clients in Asia or the Pacific I'm comfortable working async — I document clearly, communicate proactively, and don't go quiet between updates.
My Upwork track record is new. My engineering track record is not. If the project looks right, I'd genuinely like the chance to prove it.
Steps for completing your project
After purchasing the project, send requirements so Md Saidul can start the project.
Delivery time starts when Md Saidul receives requirements from you.
Md Saidul works on your project following the steps below.
Revisions may occur after the delivery date.
Discovery and scope confirmation
I review your data format, fraud types, and deployment target. I deliver a short technical spec for your sign-off before any code is written.
Data pipeline and feature engineering
I build the ingestion pipeline, preprocessing module, and feature store with your transaction schema. I validate data quality and confirm feature distributions.