Selected work
Private source
Tonalli AI — One workspace for models, tools, and follow-up
Tonalli AI brings chat, tools, files, voice, messaging, generated artifacts, and scheduled prompts into one experimental workspace.
Open live productPrivate source
Overview
I designed the interaction model, connected model and tool workflows, built persistence and messaging behavior, and tested scheduled delivery and its failure states.
Combined models, tools, artifacts, scheduling, and delivery visibility in one guest workspace.
Context
The operating problem behind the interface.
AI-assisted work becomes hard to continue when the conversation, source files, generated result, and reminder to act on it all live in different products.
Contribution
From customer signal to a system that can be used and shipped.
What I owned
I designed the interaction model, connected model and tool workflows, built persistence and messaging behavior, and tested scheduled delivery and its failure states.
What moved
- 01
I designed the model, mode, streaming-chat, and agent-tool experience around one conversation.
- 02
I connected saved chats, custom agents, memory, files, voice, messaging, and rendered artifacts.
- 03
I made scheduled delivery record its outcome and added a plain-text Telegram retry when formatted delivery fails.
Key decision
Why this decision
I persisted the delivery result, not just the schedule request. A reminder feature is only useful when the person can see whether delivery succeeded, failed, or needs another attempt.
What it taught me
Tonalli AI pushed me to think beyond the chat response. The useful unit of work includes the artifact, the next action, and a visible record of what happened afterward.
Result and current status
Result
The live demo now shows how a conversation can continue into a tool result or artifact, remain available in the workspace, and become a scheduled follow-up on the web or Telegram.
Current status
Tonalli AI is a live product demo with private source. A guest workspace is publicly viewable, while provider-dependent actions require configured services.
Evidence
Every plate is public-safe, privacy-reviewed, and labeled by evidence type. No customer or account data appears here.
01
I designed the model, mode, streaming-chat, and agent-tool experience around one conversation.
02
I connected saved chats, custom agents, memory, files, voice, messaging, and rendered artifacts.
03
I made scheduled delivery record its outcome and added a plain-text Telegram retry when formatted delivery fails.
A short synthetic flow from a conversation to an artifact and a scheduled follow-up.
Read the visual transcript
The walkthrough opens a synthetic conversation in the Tonalli AI guest workspace, shows the related artifact area, and finishes on the scheduling controls and delivery-status view. No private prompt is submitted and no provider delivery is claimed by the recording.
Limitations
- 01
The portfolio media uses synthetic guest content and does not include an account, a private prompt, or provider settings.
- 02
The public capture does not by itself confirm model availability, voice or bot delivery, or a scheduled execution.
Technology
- JavaScript SPA
- HTML
- CSS
- Vercel Functions
- Supabase
- AI providers
- Speech services
- Telegram
Source
Private source
Contact
Hiring and product teams
Need someone who can connect product judgment to delivery?
I am open to product, product operations, sales automation, and AI workflow roles where customer context matters.Discuss an opportunityFounders and operating teams