The ELEVATE server pipeline runs locally on two heterogeneous consumer machines, keeping LLM and TTS models fully resident in memory for low-latency inference¶

Claim · All claims
Evidence · 1 study · design r2 · q1
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q1 i? In the reported hardware configuration, the system ran on two local consumer machines (one for LLM generation and orchestration, one for TTS) and kept both models fully resident in memory while performing low-latency inference. → Lorenzo Stacchio 2026

Evidence¶

Lorenzo Stacchio 2026¶

Lorenzo Stacchio, Michele Giordano, Daniele Berardini, Primo Zingaretti, Emanuele Frontoni. (2026). ELEVATE: Designing Human-Centered GenAI Virtual Tutors for Scalable and Inclusive Education. arXiv preprint. https://arxiv.org/abs/2606.30662 doi:10.48550/arxiv.2606.30662

q1 · i? · design · r2

Apparatus description of the small-school deployment: the server pipeline was executed on two local machines, one running LLM generation and the Orchestrator and one running TTS, using consumer GPUs of different generations. The authors state this configuration supports "performing low-latency inference"; no latency numbers are printed in the available text.

"This configuration is capable of keeping both the LLM and TTS models fully resident in memory and performing low-latency inference (while adopting a low-cost and consumer device)."

Discussion¶

Related Claims¶

• An ELEVATE prototype achieves near real-time end-to-end tutoring interaction on consumer-grade school hardware without cloud GPUs — related
• The prototype maintains total end-to-end latency of 3-5 seconds between query completion and RAG-generated response delivery — related
• A complete four-skill open-model stack fits under 4 GB, runs at reading pace on the cheapest hardware by community measurements, and costs about one US cent of electricity per study hour — related
• Asynchronous concurrent execution bounds overall latency to approximately the slowest individual model's inference time — related