Infrastructure Engineer H/F - Algovia
Les missions du poste
What they need
Context
This initiative targets the frontier between Mixed Reality and Generative AI by modernizing how users interact with knowledge bases. The primary objective is the architectural refactoring of the existing RAG pipeline to natively support low-latency audio queries. The current setup relies on sequential, multi-step conversions (Audio Text RAG LLM Text Audio), introducing prohibitive latency and loss of context.
The objective is to redesign this architecture into a direct, low-latency streaming pipeline connecting the XR headset (Sphere/Presence framework) directly to OpenShift AI 3.5. This pipeline will process streaming audio input natively against the Elasticsearch retrieval engine and stream responses back into the user's field of view in real time.
Once the core Audio-RAG pipeline is established, the project will benchmark advanced XR-AI capabilities:
Live In-Headset Translation: Real-time audio/text translation overlay within the spatial display, and/or audio feedback.
Combining computer vision measurements with technical insights, from Audio request (e.g., measure the dimension of this crack).
Stack technique
Infrastructure: 100% On-Premise OCP (OpenShift Container Platform).
AI & Retrieval Engine: OpenShift AI 3.5 utilizing vLLM for Model-as-a-Service (MaaS) serving, interfaced with an existing Elasticsearch core.
Conditions de travail
Présent sur site les premiers jours pour intégration sinon full remote ok.
Durée jusqu'au 18/12/26 pour l'instant. La prolongation dépendra du succès de cette mission.
Pas d'habilitation particulière hormis malheureusement le classique de la nationalité française pour AH car on se fait bloquer sinon au niveau sécurité du site.
Anglais of course.
Le profil recherché
Profile wanted
Context
This initiative targets the frontier between Mixed Reality and Generative AI by modernizing how users interact with knowledge bases. The primary objective is the architectural refactoring of the existing RAG pipeline to natively support low-latency audio queries. The current setup relies on sequential, multi-step conversions (Audio Text RAG LLM Text Audio), introducing prohibitive latency and loss of context.
The objective is to redesign this architecture into a direct, low-latency streaming pipeline connecting the XR headset (Sphere/Presence framework) directly to OpenShift AI 3.5. This pipeline will process streaming audio input natively against the Elasticsearch retrieval engine and stream responses back into the user's field of view in real time.
Once the core Audio-RAG pipeline is established, the project will benchmark advanced XR-AI capabilities:
Live In-Headset Translation: Real-time audio/text translation overlay within the spatial display, and/or audio feedback.
Combining computer vision measurements with technical insights, from Audio request (e.g., measure the dimension of this crack).
Stack technique
Infrastructure: 100% On-Premise OCP (OpenShift Container Platform).
AI & Retrieval Engine: OpenShift AI 3.5 utilizing vLLM for Model-as-a-Service (MaaS) serving, interfaced with an existing Elasticsearch core.
Conditions de travail
Présent sur site les premiers jours pour intégration sinon full remote ok.
Durée jusqu'au 18/12/26 pour l'instant. La prolongation dépendra du succès de cette mission.
Pas d'habilitation particulière hormis malheureusement le classique de la nationalité française pour AH car on se fait bloquer sinon au niveau sécurité du site.
Anglais of course.
Habilitation défense