---
title: "Compression at the Edge"
category: "talks"
date: "2026-07-01"
time: "2:50pm-3:10pm"
track: "Local AI"
room: "Track 4"
speakers: ["Chris Alexiuk", "Daniel Han", "Asma Beevi", "Merve Noyan", "Parth Sareen"]
sourceLabels: ["Official conference schedule", "Public YouTube metadata"]
scheduleTrack: "Local AI"
scheduleRoom: "Track 4"
scheduleLabels: ["Local AI", "Track 4", "session", "confirmed"]
---
# Compression at the Edge

## Conference Context
- Date/time: 2026-07-01 · 2:50pm-3:10pm
- Track/room: Local AI · Track 4
- Speaker(s): Chris Alexiuk, Daniel Han, Asma Beevi, Merve Noyan, Parth Sareen
- Session type/status: session · confirmed

- Track: Local AI
- Room: Track 4
- Session type: session
- Status: confirmed

## Session Description
Compression at the Edge examines how smaller weights, faster inference, and constrained-memory deployments are making capable local AI more practical. The panel explores where compressed models already beat cloud on latency, privacy, cost, or control, what breakthroughs would unlock broader adoption, and how open model tooling is shaping the edge AI stack. Moderator: Chris Alexiuk (NVIDIA). Panelists: Daniel Han (Unsloth), Asma Beevi (NVIDIA), Merve Noyan (Hugging Face), Michael Chiang (Ollama).

## Media Evidence
[Self-Training Agents: Hermes Agent, HF Traces, Skills, MCP & Finetuning  — Merve Noyan, Hugging Face](https://www.youtube.com/watch?v=OV56RddyFuU) (speaker-match related prior/adjacent AI Engineer video; captions: English auto-captions).

- Source video: `youtube-OV56RddyFuU`
- Slide deck: [[youtube-OV56RddyFuU-dense-slides|Dense Slides: Self-Training Agents: Hermes Agent, HF Traces, Skills, MCP & Finetuning  — Merve Noyan, Hugging Face]] — 17 visible slide image(s); 17 HTML recreation(s).
![[assets/dense-slides/OV56RddyFuU/slide-001.jpg]]
![[assets/dense-slides/OV56RddyFuU/slide-002.jpg]]
![[assets/dense-slides/OV56RddyFuU/slide-003.jpg]]
- Additional slide evidence: [[youtube-OV56RddyFuU-slides|Slides: Self-Training Agents: Hermes Agent, HF Traces, Skills, MCP & Finetuning  — Merve Noyan, Hugging Face]], [[youtube-OV56RddyFuU-reconstructed-slides|Reconstructed Slides: Self-Training Agents: Hermes Agent, HF Traces, Skills, MCP & Finetuning  — Merve Noyan, Hugging Face]]
- Slide-derived themes for `youtube-OV56RddyFuU`: models, does, matter, absolute, control, over, cost, reduction.

## Evidence Graph
This evidence graph is generated from currently linked source material: official schedule text, related video pages, cached transcripts, visible slide text, dense/reconstructed slide pages, and AI slide-classification audits.

### Media Signals
- `youtube-OV56RddyFuU` — 6 slide-derived text signals
- Slide-derived themes for `youtube-OV56RddyFuU`: models, does, matter, absolute, control, over, cost, reduction.
- Evidence links for `youtube-OV56RddyFuU`: [[youtube-OV56RddyFuU]], [[youtube-OV56RddyFuU-slides]], [[youtube-OV56RddyFuU-dense-slides]], [[youtube-OV56RddyFuU-reconstructed-slides]]

### Agent Reading Notes
Use these signals to refine the synopsis, topic links, people/company context, and method notes. If a source is a related external video rather than an exact official recording, keep it framed as supporting evidence.

## Transcript Status
Related video transcript availability: English auto-captions. Treat this as supporting context, not a recording of this exact scheduled session unless later confirmed. Not fetched yet.

## People
- [[chris-alexiuk]]
- [[daniel-han]]
- [[asma-beevi]]
- [[merve-noyan]]
- [[parth-sareen]]

## Slide Evidence
- Slide-only cropped deck: [[youtube-OV56RddyFuU-dense-slides]] (20 viable slide images).
- Related slide/OCR pages:
- [[youtube-OV56RddyFuU-dense-slides]]
- [[youtube-OV56RddyFuU-reconstructed-slides]]
- [[youtube-OV56RddyFuU-slides]]
- Slide-derived terms: `model`, `skills`, `models`, `gguf`, `hermes`, `community`, `datasets`, `text`, `local`, `llama.cpp`, `inference`, `image`, `search`, `spaces`, `jobs`, `setup`, `claude`, `some`

## Synthesis
### Synthesized Breakdown
# Compression at the Edge ## Conference Context - Date/time: 2026-07-01 · 2:50pm-3:10pm - Track/room: Local AI · Track 4 - Speaker(s): Chris Alexiuk, Daniel Han, Asma Beevi, Merve Noyan, Parth Sareen - Session type/status: session · confirmed - Track: Local AI - Room: Track 4 - Session type: session - Status: confirmed ## Session Description Compression at the Edge examines how smaller weights, faster inference, and constrained-memory deployments are making capable local AI more practical. The panel explores where compressed models already beat cloud on latency, privacy, cost, or control, what breakthroughs would unlock broader adoption, and how open model tooling is shaping the edge AI stack. Moderator: Chris Alexiuk (NVIDIA). Panelists: Daniel Han (Unsloth), Asma Beevi (NVIDIA), Merve Noyan (Hugging Face), Michael Chiang (Ollama).

### Speaker And Company Context
- [[chris-alexiuk|Chris Alexiuk]] — Sr. Product Research Engineer at [[nvidia|NVIDIA]].
- [[daniel-han|Daniel Han]] — Co-founder at [[unsloth|Unsloth]].
- [[asma-beevi|Asma Beevi]] — Senior Engineer at [[nvidia|NVIDIA]].
- [[merve-noyan|Merve Noyan]] — MLE at [[hugging-face|Hugging Face]].
- [[parth-sareen|Parth Sareen]] — role not listed at [[ollama|Ollama]].

### Topics Covered
- [[agent-security]]
- [[agentic-search]]
- [[mcp]]

### Derived Links And Source Material
- [[youtube-OV56RddyFuU]] — related YouTube source page.
- [[youtube-OV56RddyFuU-slides]] — slide evidence.
- [[youtube-OV56RddyFuU-reconstructed-slides]] — slide evidence.
- [[youtube-OV56RddyFuU-dense-slides]] — slide evidence.

### Novel Concepts / Clever Methods
- No highlighted novel concept has been detected yet.

### Evidence Boundary
This synthesis is based on the official schedule and linked source pages. It should be revisited when exact session recordings or transcript-backed secondary sources are available.