In short
Remote full-stack engineer at cybersecurity company EthicalHat: build and maintain web products end to end (Java/Groovy Grails backend, Vue.js frontend, MariaDB/Neo4j) and add AI features — LLM assistants, RAG, and an MCP server that lets AI clients use product data under role-based access controls.
About the role
This is a remote position.
Requirements
•3+ years building production web applications, end to end.
•Backend: Java (OpenJDK) and Groovy with the Grails framework.
•Frontend: Vue.js, JavaScript, HTML and CSS.
•Databases: MariaDB (schema design, SQL, backup and restore with Mariabackup) and Neo4j (graph modelling, Cypher queries).
•Application server: Apache Tomcat, including TLS setup.
•Identity and access: Keycloak; SSO with OIDC and SAML using Okta, Microsoft Entra ID and Auth0; MFA and role-based access control.
•APIs: REST API design and documentation.
•Platforms: installing, upgrading and running applications on Windows Server and Linux (Ubuntu), including unattended upgrades.
•Secure coding practice: SAST scanning, dependency checks, SBOM maintenance, fixing penetration test findings.
•Clear written English for specs, API docs and user guides.
•Degree in Computer Science or similar or equivalent hands-on experience.
|
Area
|
What we expect
|
|
LLM APIs
|
Hands-on use of at least one major API (Anthropic Claude, OpenAI, Google Gemini or Azure OpenAI): streaming, tool/function calling, structured JSON output, token and cost control
|
|
Retrieval (RAG)
|
Chunking and embedding documents, vector search with pgvector, Qdrant or similar, citing sources in answers
|
|
Prompt design
|
Writing and versioning prompts, few-shot examples, system prompts that keep answers grounded
|
|
Evaluation
|
Test sets and automated checks for accuracy, hallucination and regressions before each release
|
|
AI safety
|
Guarding against prompt injection from user content, keeping customer data separate, redacting sensitive data, human approval for actions an AI suggests
|
|
Private models
|
Running open models (Llama, Mistral, Qwen) through Ollama or vLLM; switching providers behind one interface
|
|
MCP server
|
Building an MCP server with an official SDK (TypeScript, Python or Java): tools, resources and prompts; stdio and Streamable HTTP transports; OAuth-based auth; testing with MCP Inspector and real clients
|
|
Agents and frameworks
|
Connecting to external MCP servers or using agent frameworks (LangChain, LangGraph, Spring AI, LlamaIndex) where useful
|
|
Observability
|
Logging every AI call with inputs, outputs, latency and cost; alerts on failures
|