Top AI Demos #28: Real-Time Audio Classification, Federated Learning Privacy, and Interactive Presentations

Top AI Demos #28: Real-Time Audio Classification, Federated Learning Privacy, and Interactive Presentations

Issue #28 · Week of May 25

Joe Heitzeberg • Founder at AI Tinkerers • ⏱️ 1 min read

Creating space for leading builders to share ideas, grow, and make an impact.

AI Tinkerers this week were busy shipping concrete technical systems, from privacy-preserving federated learning like Addalin Lariena Rose Devasahaya Rajan’s work in Doha to real-time audio classification pipelines on Azure by Arlene Riona, also in Doha.

We’re also seeing clever multi-agent orchestration and control, like Chengkun Li’s MuscleMimic for embodied AI, and Dan Noskin’s Reflect for live AI-driven UI editing.

Several builders are also tackling developer tooling, including Bruno Perez’s Manifest for low-cost LLM routing and Ilia Azizi’s BegooAI for interactive presentations.

BegooAI: Interactive Presentation Companion Ilia Azizi PhD Candidate at UNIL 📍 AI Tinkerers - Geneva • May 13 Cooking Read more → TECH STACK Loading tech tags... PROJECT LINKS begoo.ai youtu.be Video @ilia_azizi

Privacy in Federated Learning Addalin Lariena Rose Devasahaya Rajan AI Engineer at Global Identity for Technology Services 📍 AI Tinkerers - Doha • May 18 Cooking Read more → TECH STACK Loading tech tags... PROJECT LINKS github.com Video

Real-Time 911 Audio Classification with Azure Arlene Riona Research Assistant in Affective Computing & AI at University of Doha for Science and Technology 📍 AI Tinkerers - Doha • May 18 Cooking Read more → TECH STACK Loading tech tags... PROJECT LINKS github.com linkedin.com

MuscleMimic: Scalable Muscle-Actuated Imitation Learning for Embodied NeuroAI Chengkun Li Phd student at EPFL 📍 AI Tinkerers - Geneva • May 13 Cooking Read more → TECH STACK Loading tech tags... PROJECT LINKS github.com epfl.ch Video ArXiv Paper @ckli85

Reflect: UI on the Fly Dan Noskin Founder at Ceramic 📍 AI Tinkerers - New York City • May 13 Cooking Read more → TECH STACK Loading tech tags... PROJECT LINKS reflect.ooo @dnosk

More Great Builds

Quick hits from the community — demos worth bookmarking:

Differ: Per-User App Versions

Noam Tenne • AI Tinkerers - New York City • May 13

Noam Tenne from Sky Valley presented Differ, a deployment platform where each user gets their own version of the app, and the demo showed how a single project could route different experiences based on real user actions. Differ runs by treating release like a build farm, with routing on top to resolve the right stored artifacts per user identity. Noam’s backend and DevOps background in Docker and Kubernetes showed up in the way the system was explained as build versus runtime. We liked it because it makes per-user adaptive UI practical, and people seemed to enjoy that the distribution and update assumptions finally got challenged, which is timely for the agentic, long-horizon world.

Loading tech tags...

getdiffer.comVideo

Manifest: Low-Cost LLM Routing

Bruno Perez • AI Tinkerers - Geneva • May 13

Bruno Perez presented Manifest, an open-source LLM router that smartly directs each agent query to the right model to cut inference costs. Manifest routes by complexity, specificity, and custom HTTP headers, and it lets you mix providers including local models, subscriptions, and API keys, with per-request fallback when something fails. People seemed especially into the dollar-level tracking, limits, and notifications, since it keeps multi-model orchestration practical for real traffic. We liked it because it made cost control and reliability feel like everyday infrastructure rather than a research project.

Loading tech tags...

manifest.buildgithub.com

FolioRankAI: Personalized Photo Selection

Ethel Zhang • AI Tinkerers - New York City • May 13

Ethel Zhang presented FolioRankAI, a local-first smart photo selection engine that ranks your shots by both eye appeal and technical quality like composition and exposure. It runs an orchestration of open models on-device, adding smart eye detection for closed eyes and grouping duplicate images so gigabytes of photos don’t need to leave your machine. With its privacy-focused “zero user data” approach, it also hinted at why people found it immediately useful. We liked it because it turned personalization and heavy data handling into something builders could copy and ship.

Loading tech tags...

bringezback.comapple.comVideo

LuxPricer: Market Intelligence Report

Sonya P. • AI Tinkerers - Geneva • May 13

Sonya P. built and demoed LuxPricer, an AI pricing report that scans live listings across resale marketplaces after you upload a photo, then returns a price in seconds. Under the hood, it leans on Playwright-driven browser automation to consistently collect cross-site signals and compile them into one valuation. Because it felt immediately useful for real-world market research (people seemed to love it), it stood out as agentic web navigation that can become a practical product for anyone tracking price moves and demand.

Loading tech tags...

luxpricer.comapple.com

Presentations Made Easy

Muhammad Khas • AI Tinkerers - Doha • May 18

Muhammad Khas from Scale AI spotlighted Presentations Made Easy, an internal slide generator that takes a slide title plus a few details and produces on-brand slides with built-in interactivity. The tool focuses on tightening the “last mile” between raw AI output and something you can actually reuse, using a repeatable generation workflow rather than one-off prompting. It fit the community pain point perfectly, and the room feedback leaned toward “this finally saves time without breaking consistency.” We liked it because it turns long-horizon work into a dependable asset for everyday teams.

Loading tech tags...

You are one of 95,000+ readers from Apple, Amazon, Microsoft, Google, Nvidia, OpenAI, Anthropic, Cohere, ElevenLabs, Scale AI, Groq, Mistral AI, and others — spanning frontier labs, big tech, startups, and top universities.