View all jobs

LLM Develper (JARVIS) - September 2026

  • remote, remote
For one of our clients in the pharma industry, we are looking for an LLM Developer (JARVIS)

Project Name: JARVIS – Meta-Agentic AI Framework Development (MVP Phase)

Project Description: The service is requested as part of the JARVIS project. JARVIS is a meta-agentic AI framework developed by the client’s X Digital Innovation Lab that creates custom agentic workforces for individual business domains within the client. The project is currently entering the MVP phase, focusing on training and testing JARVIS across targeted real-world use cases. The objective of this assignment is independent technical engineering, architectural design, and quality validation of AI-produced outputs to deliver a validated meta-pattern and working prototype agents. The goal is to provide technical evidence regarding desirability, viability, and technical feasibility to support a future go/no-go pilot decision.

Tasks:
1.Technical design and conceptualization of JARVIS components
  • Drafting of technical concepts and implementation plans covering agentic process/pipeline design, interactions with the Apollo AI gateway, tool and Model Context Protocol (MCP) boundaries, operational constraints, and quality-gate validation approaches.
2.Technical programming of LLM-enabled Python services
  • Independent execution of programming tasks to develop agentic workflows utilizing the client's Apollo AI gateway.
  • Technical integration of AI coding-assistant tooling, asynchronous/streaming processing, OAuth2 authentication, configuration handling, and run-mode management.
3.Development of agentic workflow architectures
  • Technical programming and structuring of core workflow components, including tool invocation, function calling, Model Context Protocol (MCP) integration, state handling, and error handling.
  • Drafting and technical configuration of multi-stage pipeline orchestration models (deliver → expert → review → quality-gate).
4.Technical development of web interaction interfaces
  • Execution of programming tasks to create frontend interaction surfaces for JARVIS workflows, specifically utilizing Streamlit.
  • Technical implementation of input forms, result and status views, feedback capture mechanisms, and usage/budget displays.
5.Technical implementation of observability and governance mechanisms
  • Drafting and programming of evaluation and guardrail mechanisms for agentic applications.
  • Technical configuration of cost and token tracking, hard budget caps, latency tracking, and output-quality monitoring gates based on defined use-case parameters.
6.Containerization and deployment configuration
  • Execution of containerization tasks utilizing Docker and OpenShift (client's internal OpenDevStack/EDP platform).
  • Technical drafting of Kubernetes/OpenShift configurations, CI/CD pipeline definitions, and SSO-gated environment parameters.
7.Execution of technical quality assurance and testing
  • Independent execution of testing protocols on implemented backend, workflow, and interaction components.
  • Technical expansion of existing automated suites including unit tests, integration tests, and browser-based E2E scenarios.
  • Drafting of structured defect reports documenting identified high-severity anomalies for handover to client project governance.
8.Creation of technical documentation and handover reports
  • Drafting of comprehensive technical documentation covering architecture decisions, output results, operational monitoring setup, and known limitations.
  • Creation of a finalized project report consolidating all implemented JARVIS components and evaluating the extracted meta-patterns.
Deliverables: Creation of comprehensive documentation with all results regarding the above-mentioned tasks—including agentic process and pipeline modules, workflow and interaction components, evaluation assets, an OpenShift deployment package, and test documentation—with subsequent handover to client for review and approval for further usage.


Skill Requirements:
  • Senior-level consultant (5+ years) is recommended. JARVIS is entering a fast-moving MVP phase with high ambiguity and no established playbook, so the contractor must independently make architecture and retrieval-strategy choices with minimal onboarding. Beyond 3+ years of hands-on LLM/RAG/agentic development experience to reduce ramp-up risk, the person should demonstrate strong ability to communicate and decompose complex problems into simple, actionable solutions.
  • Experience implementing agentic systems in a pharma/healthcare context is a plus, though not mandatory. The contractor should also be able to mentor less experienced team members and show a strong ability to keep up with the fast-moving learning curve of AI tooling, models, and emerging technology.
  • Pharma/healthcare/regulated-environment experience (GxP, GDPR, EU AI Act, computerized system validation) is a nice-to-have, not a must-have. Prior exposure to a regulated or data-sensitive environment is a plus and genuine curiosity to learn responsible-AI practices in pharma is an acceptable substitute.
  • AI coding assistants and LLM tooling — GitHub Copilot in particular — and prompt engineering, used through an enterprise AI gateway (Apollo)
  • Agentic workflow orchestration and Model Context Protocol (MCP) tool integration; experience with orchestration frameworks (e.g. LangGraph, CrewAI) is transferable
  • Python as the primary language; the current frontend uses Streamlit, with the frontend approach open to evolve during the MVP (React/TypeScript beneficial for prototyping)
  • Docker, Kubernetes/OpenShift, and CI/CD, deploying on client’s internal OpenShift (OpenDevStack/EDP)

Nice to have:
  • experience running local LLMs on dedicated hardware; familiarity with Dify (relevant to Apollo Studio).
  • Familiarity with responsible-AI/security practices such as the OWASP Top 10 for LLM Applications and basic bias/fairness evaluation is a nice-to-have, and
  • relevant given JARVIS's data-confidentiality and governance-alignment challenges.

Language requirements:
  • Fluent English is mandatory, as it is the working language of the team and stakeholders. German is optional/beneficial but not required.

Start: ASAP
Capacity: 25h/week
Duration: till end of February 2027
Location: 80% remote, 20% onsite (Ingelheim am Rhein)
Hourly rate: 78 Euro/h