Forward Deployed Software Engineer
Forward deployed: I embed with your team, speak both business and tech, and ship production software on Clean Code and SOLID principles. Six-plus years of enterprise delivery in regulated industries. I integrate GenAI where it counts, with LLM features that run on-prem and keep your data in-house.
Start a conversation →Forward Deployed Engineer
I sit with the business side, learn the domain, and turn the real problem into shipped software. No requirements thrown over a wall.
GenAI Integration
I bring LLMs into production where they concretely help and lead teams through the adoption curve. A realistic read on what they can and can't do.
Software Engineer
I build complete systems: Java/Kotlin or Python on the backend, Vue 3 or Angular on the frontend, Docker and OpenShift in the infrastructure. Clean, maintainable code on Clean Code and SOLID principles.
Privacy-first AI
I run AI on-prem with local inference, so sensitive data never leaves your infrastructure. Built for GDPR and regulated environments.
I'm a software engineer who builds with AI tools daily and ships them in production. That means a realistic read on what LLMs can and can't do, clean integration with existing systems, and the judgment to keep sensitive data on-prem.
From requirement to production: architecture, implementation, deployment. Full stack or focused on backend or frontend, depending on the gap. Maintainability and scalability built in from day one.
Pay down technical debt, modernize architecture, integrate legacy systems. I know enterprise stacks and how to protect existing investments while moving forward.
Assess where GenAI brings real value in an existing system. Implement LLM features: local or cloud-hosted, privacy-compliant, embedded in existing workflows. Not bolted on. Real integration.
CI/CD pipelines, quality gates, deployment automation. I build toolchains that make teams permanently faster and safer. Already established as the delivery standard for complex enterprise accounts.
Full-stack development, AI integration, productization: from idea to government grant
What we built
A time-series-native observability platform (Vue 3 / Python / FastAPI) with configurable widgets, FQDN-based data sources, BFF auth pattern, and a no-code early-warning system.
AI component
An autoencoder trained on customer data spots anomalies and triggers incident triage. A multi-agent LLM system classifies the incidents and produces traceable explanations. A model-agnostic gateway runs local inference: privacy-compliant, with no changes to application logic.
Outcome
Production-ready platform, EXIST federal grant approved. Technical responsibility within the founding team for architecture, backend, frontend, auth, and AI integration.
Digital workspace for mid-market property underwriters and international team collaboration
Starting point
Underwriters worked across distributed systems with no straight-through integration. Local and international teams coordinated through manual handoffs. Rekeying between systems cost time and produced errors. The focus on the actual complexity of policy work got lost.
What I built
An integrated underwriting workbench. From requirements work with the business side through to production. System jumps eliminated through unified workflow integration. Shared workspace for local and international teams, embedded in existing underwriting processes without parallel structures.
Outcome
Underwriters can focus on the actual complexity of policy work instead of on system switches. Smooth collaboration across sites. Full delivery responsibility: requirements, architecture, implementation, go-live.
Microservice delivery for a reinsurance client in a regulated environment
Starting point
No automated delivery process. Releases were manual, inconsistent, and error-prone. The project had to build engineering discipline while shipping at the same time.
What I built
An end-to-end pipeline from SemVer hooks through CI builds and an artifact registry to container deployment, with quality and security gates wired in. Scalable microservice architecture with message queues, API gateways, and identity management.
Outcome
Toolchain adopted as the standard for every follow-up project. Promoted to Lead Engineer after eighteen months. Business-rules engine integrated, business team enabled to extend it on their own.
Migration prep for a grown database landscape in a regulated environment
Starting point
A grown Oracle database landscape was set for cloud migration. Unclear: which objects were migration-critical, where the data model needed reworking.
What I built
A term-frequency matrix across DDL and PL/SQL source code that surfaces dependencies and usage frequency. On top of that: a joint reassessment of the data model with the domain team.
Outcome
Clear prioritization of migration-critical objects, a reworked data model as the foundation. Not lift-and-shift with old debt, but a structured restart in the cloud.
SQL, ETL, data modelling, and reporting in Cognos and Tableau
Question
Which existing drugs interact with a vaccine in the development phase? The answer was in the data. It just had to be made structured and queryable.
What I contributed
Wrote the SQL queries that surfaced relevant interaction patterns from Oracle source data. Designed the ETL pipelines and the underlying data model. Built the reporting environment in Cognos 11 and Tableau for the domain team.
Context
Drug safety data in a regulated pharma environment: precise queries and a clean data model are not optional, they are the prerequisite. The work required close coordination with the domain experts to translate the right questions into database logic.
This site (architecture, components, i18n scaffolding, A11y pass, mobile navigation) was built with AI assistance during a single lunch break. The tools write code. The architecture, the trade-off calls, the question of what to build at all, and whether the code is clean and maintainable on Clean Code and SOLID principles: that stays human work. Six years of enterprise delivery before this current tool cycle make the point: I deliver without AI too. With AI I'm just faster.
At these speeds the failure mode is not slow code. It is fragmentation. Context drifts across dozens of AI-generated files nobody fully read. Specs written before the first token is generated are the countermeasure. They hold the architecture together when every change takes minutes instead of days.
Federal grant · 2025
Competitive German federal grant for technology-driven startups. Reviews technical feasibility and commercial viability.
Enterprise IT consultancy · Promotion 2023
Promoted after eighteen months at one of Europe's largest IT consultancies. Architecture responsibility, team coaching, engineering standards established at account level.
B.Sc. · FOM Frankfurt · 2024
Bachelor thesis on labelling AI-generated content in social media. Technical and regulatory perspectives joined up.
AWS + Blockchain · 2020 / 2022
Certified foundations in cloud architecture (AWS) and distributed systems (Ethereum).
Highlighted = focus area · Grey = used in production
Whether the project is a new build, a modernization, or AI integration, I am happy to start with a conversation.
mail@dionjones.com →