Technologies
The stack behind our work.
We keep the list short on purpose: proven technologies we have shipped with, know how to operate and can hand to your team. Models are chosen per product after benchmarking on your own data.
Reviewed against current demand on 11 October 2026 · revisited each quarter
AI and machine learning
Models, orchestration and measurement for the AI-native track. We benchmark on your own data before we commit to a model, and every call is traced.
- Claude
- Anthropic models through the Claude API and Agent SDK for agents, long documents and tool use.
- Gemini and Vertex AI
- Gemini models, Vertex AI Agent Engine, Vector Search and managed training on Google Cloud.
- OpenAI
- GPT models, the Responses API and Agents SDK where they fit the task.
- Open-weight models
- Llama, Mistral, Qwen and Gemma families for fine-tuning and private deployment.
- Python
- The language of the model work: training, evaluation, serving and data pipelines.
- LangChain and LangGraph
- Retrieval chains and tool use, and stateful, durable agent orchestration with checkpoints and human-approval steps.
- Spring AI
- Chat clients, tool calling, vector stores and MCP inside Spring Boot, so AI-native CRMs share the platform's backend.
- Model Context Protocol
- Standard tool interfaces so agents reach your systems through one audited door.
- PyTorch and Hugging Face Transformers
- Training, LoRA and QLoRA fine-tuning with PEFT, and evaluation of custom models.
- vLLM
- High-throughput inference serving on your own GPUs.
- Vector databases
- Qdrant for filtering, hybrid search and scale; MySQL 9 VECTOR, HeatWave Vector Store or pgvector when the relational store is enough.
- Langfuse and OpenTelemetry
- Traces, cost per request and evaluations, self-hosted and vendor-neutral.
- Ragas and promptfoo
- Retrieval and prompt evaluation harnesses with a regression run on every change.
Full-stack Java
The traditional track. Systems with real business rules and long lifetimes get Java and Spring Boot: typed, tested, containerised and maintainable after us.
- Java 21 and 25 LTS
- Current long-term-support releases: virtual threads, records, pattern matching.
- Spring Boot 3 and 4
- Our default for CRMs, platforms, APIs and integrations, on Spring Framework 6 and 7.
- Spring Security and Keycloak
- Authentication, authorisation, roles, OAuth 2, OIDC and single sign-on.
- Spring Data JPA, Hibernate and Flyway
- Persistence with versioned migrations.
- Spring Modulith
- Well-bounded modules inside one deployable before anyone reaches for microservices.
- Apache Kafka
- Event streaming between services and systems, with Pub/Sub on Google Cloud where simpler.
- REST, OpenAPI and GraphQL
- Documented contracts that front ends and partners can build against.
- JUnit 5 and Testcontainers
- Integration tests against real MySQL, Kafka and Redis in CI.
- React and Next.js front ends
- The screens on top of a Spring Boot backend.
Data
MySQL is our default relational database, it runs managed on both clouds we work in, and vector search now lives inside it.
- MySQL 8 and 9
- Schema design, indexing, replication and query tuning, with the VECTOR type for embeddings next to business data.
- Cloud SQL for MySQL
- Managed MySQL on Google Cloud with backups and high availability.
- OCI MySQL HeatWave
- Managed MySQL on Oracle Cloud with in-memory analytics, Vector Store and GenAI.
- Oracle Database 23ai
- Autonomous Database with AI Vector Search for enterprise workloads.
- Qdrant
- A dedicated vector database when filtering, hybrid search or scale outgrow the relational store; pgvector where you already run PostgreSQL.
- Redis
- Caching, rate limits, sessions and queues.
- BigQuery and Firestore
- Analytics and feature data for training; document storage for web and mobile products.
Cloud and DevOps
We build in your Google Cloud or Oracle OCI account, as code, with CI/CD and observability from the first week.
- Google Cloud
- Cloud Run, GKE, Cloud SQL, Vertex AI, Pub/Sub, Cloud Storage and Firebase.
- Oracle Cloud Infrastructure
- Compute and GPU shapes, OKE, MySQL HeatWave, Autonomous Database, Object Storage and OCI Generative AI.
- Kubernetes
- GKE and OKE for services that outgrow serverless, with Cloud Run first where it fits.
- Terraform and OpenTofu
- Every environment described as code and reviewed in pull requests.
- Docker
- One image per service, the same locally and in production.
- GitHub Actions and Cloud Build
- Tests, builds and deployments on every merge.
- OpenTelemetry and Grafana
- Traces, metrics and logs with Cloud Monitoring or Grafana on top.
- Vercel
- Hosting for Next.js front ends and marketing sites.
Web
Front ends that load fast, pass accessibility checks and can be read by search engines and AI assistants.
- Next.js 16
- App Router, server components, server actions and the Metadata API.
- React 19 and TypeScript
- Typed components and a small shared design system.
- Tailwind CSS 4
- Consistent styling without a large CSS codebase.
- Playwright
- End-to-end tests that run in CI against a real browser.
- PostHog
- Product analytics, session replay and feature flags.
Desktop and mobile
The stacks behind ByteMirror, ByteLoader and ByteViewer.
- Swift 6 and SwiftUI
- Native iOS and iPadOS apps, widgets and Shortcuts.
- Rust and Tauri 2
- Small, fast Windows desktop apps with a web interface.
- Device protocols
- USB, Wi-Fi and Apple device services for installing and streaming.
- Hardware video decoding
- H.264 and H.265 at 1080p60 on the GPU.
FAQ
About our stack
Which versions are you on right now?
Java 21 or 25 LTS with Spring Boot 3 or 4, MySQL 8 or 9, Next.js 16 and React 19, Tauri 2 and Swift 6. We review this list each quarter and move when the LTS release and the ecosystem are ready, not on release day.
Why Java Spring Boot and MySQL for backends?
They are the safest choice for systems that must run for years: strong typing, a mature security model, predictable performance, managed hosting on both Google Cloud and Oracle OCI, and a large pool of engineers who can maintain them after us.
How do you orchestrate and measure agents?
Python with LangChain for retrieval chains and LangGraph for stateful agents with checkpoints and human-approval steps, Spring AI when the agent lives inside a Spring Boot platform, the Model Context Protocol for tool access, Qdrant or MySQL 9 for vector search, and Langfuse with OpenTelemetry for traces, cost per request and evaluations. Ragas and promptfoo run the retrieval and prompt regression suites.
Which AI models and APIs do you work with?
Claude through the Claude API, Gemini on Vertex AI, OpenAI, and open-weight families such as Llama, Mistral and Qwen for fine-tuning and private deployment. We pick per product after benchmarking on your own test set.
Can you work with a technology that is not listed here?
Often yes, especially for integrations. This page lists what we use by default and have shipped with. If your system is built on something else, ask and we will say plainly whether we are the right fit.
Do you use AI tools in your own engineering?
Yes. We use AI coding assistants for drafting, review and tests, with every change reviewed by an engineer. Our products and this site are built that way.
See it applied
Each service page lists the exact stack it uses.
Tell us the job. We'll tell you whether it needs AI.
A short call, a written brief, and a straight answer on scope and cost before any work starts. NDA first if you need one.