# Solving the Memory Wall: A Deep Dive into AI Inference with Sandra Rivera

Page: https://stenobird.com/podcast/amelia-s-weekly-fish-fry-166630/solving-the-memory-wall-a-deep-dive-into-ai-inference-with-sandra-rivera
Text version: https://stenobird.com/podcast/amelia-s-weekly-fish-fry-166630/solving-the-memory-wall-a-deep-dive-into-ai-inference-with-sandra-rivera.md
Podcast: [Amelia's Weekly Fish Fry](https://stenobird.com/podcast/amelia-s-weekly-fish-fry-166630)
Published: 2026-04-10T07:00:00+00:00
Episode link: https://techfocus.podbean.com/e/inside-visora-how-a-french-startup-is-rewiring-ai-inference/
Audio file: https://mcdn.podbean.com/mf/web/fy9wdqmrhqdmm8dy/FF_-_April_10_take_16gun5.mp3
Processing state: not_requested
JSON: https://stenobird.com/v1/public/podcasts/amelia-s-weekly-fish-fry-166630/episodes/solving-the-memory-wall-a-deep-dive-into-ai-inference-with-sandra-rivera
Duration seconds: 1000

## Resource

This week, I'm excited to welcome Sandra Rivera from VSORA! We dive into a discussion on why AI inference is essential for deployment at scale, specifically focusing on how VSORA’s patented software architecture addresses the "memory wall" by collapsing memory layers. We explore their recent tape-out, which promises approximately 3X the performance at half the power of leading GPUs. We also chat about deployment use cases, the need for low latency and high determinism, future plans for OEM modules and MLPerf benchmarking, and even get a brief look into Sandra’s family llama farm.

## Actions

- request_transcript: `POST https://stenobird.com/v1/public/podcasts/amelia-s-weekly-fish-fry-166630/episodes/solving-the-memory-wall-a-deep-dive-into-ai-inference-with-sandra-rivera/transcription-requests` — Idempotently request low-priority transcript generation for this episode.
- read_markdown: `GET https://stenobird.com/podcast/amelia-s-weekly-fish-fry-166630/solving-the-memory-wall-a-deep-dive-into-ai-inference-with-sandra-rivera.md` — Read the agent-friendly Markdown representation of this episode resource.

A page view does not enqueue transcription. Agents should invoke `request_transcript` explicitly when they need this episode processed.

## Transcript

Full transcripts are not published on public pages unless there is a clear rights basis.