Client names are withheld under confidentiality. I can go into detail in a call.
2026.03 — Present / Public sector / 7 engineers, 30 people total
Operations and maintenance of a budget and revenue/expenditure management system for local government
A system used by the finance division of a local government to allocate revenue against expenditure and to manage project budgets across fiscal years. The system was already in service; I handle its operations and maintenance, including new features and bug fixes, covering design through implementation and integration and system testing.
Built an admin tool that reconciles and joins CSV files spanning multiple fiscal years, delegating the choice of join keys and join order to Amazon Bedrock. Only metadata — row counts, columns guaranteed to exist — goes to the model, and the result comes back as structured output against a JSON Schema, so it can be validated rather than trusted. I also migrated the DRF-based API to Django Ninja + Pydantic, and implemented exclusive control as a decorator other jobs could adopt, so that Celery batches firing every ten minutes never run twice across tenants.
Python / Django / Django Ninja / Celery / TypeScript / React / PostgreSQL / Amazon Bedrock / Datadog
2025.07 — 2026.02 / Information and telecommunications / 7 engineers, 30 people total
Migration and feature work on a sales support system for mobile phone stores
A sales support system where store staff register visiting customers from a tablet and use that data in their sales activity. The existing system had aged and could no longer be extended, so it was being moved onto a new platform. I covered detailed design through system testing.
Reworked screens and added features with Lambda (Node.js) and Vue.js, including CSV aggregation and download, changes to permission handling at sign-in, and per-screen display control. Ahead of a large release I wrote and ran tests from the system-test checklist and cleared the resulting display-control defects. Where paging through customer detail screens multiplied API calls, I verified the behaviour with performance testing.
TypeScript / Vue.js / Node.js / PostgreSQL / Lambda / ECS Fargate / Terraform
2024.11 — 2025.06 / Information and telecommunications / 10 engineers, 12 people total
Matching system for consulting talent
A matching platform running between three parties: the broker, consulting firms, and companies that need consultants. When an engagement is registered, the broker either puts out a call to the firms or proposes candidates using generative AI; the flow then runs through interviews and a hiring decision to registering the contract. I covered detailed design through system testing.
Owned design and implementation across the front end (Nuxt.js) and back end (Django), and ran the local development environment. Implemented on-demand CSV aggregation and download, plus scheduled aggregation with upload to S3, on Lambda. Automated part of the scenario testing with Playwright MCP.
TypeScript / Nuxt.js / Python / Django / MySQL / ECS Fargate / SQS / Docker
2023.12 — 2024.09 / Electrical equipment / Lead, team of 4
A question answering system over internal documents
A proof of concept for an AI assistant that answers questions about the documents used inside the company, built as RAG — a large language model combined with retrieval. Run as an agile project; as lead I owned it from planning and requirements onward.
Designed the RAG architecture and the AWS setup myself, built the answer-generation API with FastAPI and the UI with Next.js, and used Pinecone as the vector database. Set up CI/CD with CodeBuild, CodeDeploy, and CodePipeline through Terraform, and handled code review.
TypeScript / Next.js / Python / FastAPI / LangChain / Pinecone / Terraform / ECS Fargate
2023.09 — 2023.11 / Construction / Team of 3
An evaluation harness for a question answering system
A question answering system had been built to strengthen knowledge sharing from internal manuals and help train junior staff. Our scope was choosing the evaluation libraries and designing the evaluation process. I started from requirements interviews.
Surveyed the available evaluation libraries, then implemented an evaluation process using SentenceBERT and BERTScore. The point of the project was to replace "it feels better now" with numbers you can compare; it runs on Azure OpenAI with Azure App Service and Functions.
Python / Transformers / Azure OpenAI / Azure App Service / Docker
2023.07 — 2023.08 / Manufacturing / Lead, team of 2
Requirements interviews and estimation for an AI parameter tuning system
On a project automating parameter tuning with AI for a manufacturer, drawing out the requirements for additional features and producing an estimate. As lead I handled the discussions with the commissioning side.
Ran the requirements interviews and the technical investigation, then presented what was feasible and what it would cost as a written estimate.
Python / Django / Node.js / React / SQLite / Docker
2023.04 — 2023.06 / Retail / Team of 3
A demand forecasting proof of concept with time-series models
A proof of concept testing whether time-series models could forecast daily demand for sporting goods well enough to improve inventory management. I covered requirements through implementation.
Started from surveying the techniques and designing the database that would hold the data, then implemented the statistical models and owned the results and their interpretation. Training was too slow locally and needed a GPU, so we ran it on AWS SageMaker; I also set up accounts and roles for the team and managed the schedule.
Python / PyTorch / statsmodels / scikit-learn / MySQL / AWS SageMaker / Docker
2022.12 — 2023.03 / Internal training / Lead, solo
A CRUD training site for internal onboarding
The company needed a site where new hires could learn basic CRUD operations. I ran it alone, from planning and requirements through design, implementation, testing, operations, and support.
Designed the architecture and wrote the requirements document, feature inventory, and test specification myself, then implemented the CRUD features in the shape of a small e-commerce site. Built with Django and React, provisioned on Fargate, RDS, CodeBuild, and CodePipeline through Terraform.
Python / Django / TypeScript / React / MySQL / ECS Fargate / RDS / Terraform / Docker
2022.08 — 2022.11 / Manufacturing / Team of 3
Front end, API, and back end for an AI parameter tuning system
Parameter settings for the machinery were being entered by hand, which concentrated the load on the few people who knew how. The project built a web application that automates that work with AI; I implemented it from the screens through to the back end.
Built the API and back end with Django and the screens with React, and put the back-end unit tests and test code in place.
Python / Django / TypeScript / React / PyTorch / SQLite / Docker
2021.04 — 2022.06 / Retail / Sub-lead, team of 6
Demand forecasting system for retail stores
Sales and inventory management at the retailers were still manual. This in-house product forecast sales with AI so that inventory management could be automated and merchandising decisions sharpened. Run as an agile project; as sub-lead I covered planning through system testing.
Interviewed the retailers we worked with about the screens and features they needed, wrote the requirements document, and took it through screen design, the feature inventory, and database design. Designed the AI pipeline architecture and implemented it in Python, provisioned the AWS setup (EC2, RDS, ECS, Lambda, Step Functions) with Terraform, and reviewed front-end and back-end work.
Java / Spring Boot / TypeScript / React / Python / MySQL / AWS / Terraform
2021.02 — 2021.03 / Food / Team of 5
Testing for feature additions to a customer and product management system
Adding features to an existing customer and product management system to make the work it supports more efficient. I was responsible for integration testing.
Worked as the test engineer on integration testing for the added features, and took part in setting up the environment on Docker and EC2.
Java / Spring Boot / TypeScript / Vue.js / MySQL / EC2 / Docker