❯ demos
Public demos play right here. Demos of Emilia products running at real customers are private: their descriptions are public and the videos are shared personally, on request.
32 demos · 11 public videos · 20 private
Products I designed at Emilia AI, from proof of concept to production. The videos show real customer deployments, so they are private.
Loss prevention at the counter: a virtual tripwire on the camera detects customer–cashier exchanges, groups them into sessions and reconciles them with the POS. Exchanges with no recorded sale go to human review.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Pay-by-weight buffet: every plate that goes on the scale is detected and matched to its POS order. Plates with no matching sale are flagged for review.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Pay-by-weight buffet with two scales watched by a single camera. Every plate is reconciled with the POS.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Branded cups are tracked on video and matched to fiscal receipts captured from the printer. Anomalies are flagged both ways: cups without a sale and sales without a cup.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Revenue assurance for a fixed-price-per-person restaurant: people seated at each table are counted periodically and compared with the admissions charged. Differences are evidence to review, not accusations.
RoleI designed the architecture and built the proof of concept.
A physical store's funnel: interactions of interest seen by several cameras against recorded sales. Every sale opens with its visual evidence and its fiscal receipt.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Events detected by a virtual tripwire are scored with open-vocabulary segmentation to find trays. A human review queue turns every decision into training labels.
RoleI designed the architecture and built the proof of concept.
Production review at a pizzeria: every pizza detected, flavor analysis constrained to the menu, a synthetic kitchen ticket and visual menu references.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Every delivery box is detected and described (dish, packaging, confidence) and summarized in a synthetic ticket to compare with what was ordered.
RoleI designed the architecture and built the proof of concept.
Delivery orders verified line by line on video of the packing bench, including the kits the customer asked for or declined. Every check is timestamped on the video.
RoleI designed the architecture and built the proof of concept.
Counter microphone, two cameras and POS on one clock. Every sale is segmented into arrival, order, payment and close, with its transcript.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
A synchronized replay of every sale with its POS record, and service coaching (offer water, explain kindly, suggest an add-on) backed by the transcript.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
A daily service scorecard by day and by operator, with an explainable analysis of every session and an analytics view.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Audio, register and video on one clock: every payment moment is classified by method (cash, instant payment or card) and by the source that observed it. Differences are reported conservatively for review.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Dispatch tickets are reprinted with our own QR code; a camera on the pass detects it and records the moment each order is ready for pickup.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Every entry into the stock room is logged with time and dwell duration, one click away from the exact moment on video.
RoleI designed the architecture and built the proof of concept.
At a quick-service counter, every dish is followed from the moment it leaves the kitchen, enters the pass, is called and is picked up, with service-time metrics.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Using the camera the store already has, every freezer opening and closing is logged with its duration; long cycles are highlighted together with an estimate of thermal exposure and energy use, with video review.
RoleI designed the architecture and built the proof of concept.
A diverse queue of served dishes where the chef describes, in plain language, what should be measured in each one.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Structured expert review (bake, toppings, bubbles, fit in the box) saved as validated JSON.
RoleI designed the architecture, built the proof of concept and took it to production end to end, with the Emilia team.
Coding-agent tools, conversational voice AI and generative AI experiments, most of them first shared on X.
A maintained fork of Nico Bailon's visual-explainer skill, which turns complex terminal output into HTML pages. It syncs with the original project and applies declarative preferences that the agent interprets with judgment, instead of mechanical patches.
RoleFork author
An explainer video about Pi Durable, a project by Earendil (pi.dev). The project isn't mine: I made the explainer, produced with the pi agent itself, manim animations and a Kokoro voice.
RoleAuthor of the explainer video; Pi Durable is an Earendil project
A RAG assistant to support professional career development.
RoleLed development as CTO of RisePal
A real-time English voice tutor from ArtisanLabs. In the demo it practices Walt Whitman's vocabulary and switches language when asked.
RoleAuthor (solo project)
A voice assistant with RAG to discuss the architecture of any application, built for the Supabase hackathon.
RoleAuthor (solo project)
A voice agent on the open-source Vocode framework that tells the story of Coqui's shutdown, summarizes its team's farewell and points people to Coqui TTS/XTTS. It follows Vocode's YC Startup Advice Bot demo.
RoleAuthor
A preliminary benchmark of time to first token and tokens per second for conversational AI, comparing GPT-3.5 Turbo with GPT-4o mini on its release day.
RoleAuthor
An early experiment, weeks after Stable Diffusion's release: my own profile portraits generated with the model.
RoleAuthor
A visual journey through Jimi Hendrix's music, with Midjourney V5.2 images, animation in Runway and editing in Kdenlive.
RoleAuthor
An NVIDIA Jetson Nano as home infrastructure for experimenting with AI.
RoleAuthor