Dash Agent

• 

  The Architecture

Inspiration

I wanted to buy a gift for someone and ran into some trouble with shipping and had to navigate many pages to solve it. I realized many people probably experience similar things and there is a need for agentic flow in consumer affairs.

What it does

It can remember and keep state and act on a users behalf. It uses a live Playwright-driven browser, Gemini's reasoning, and a network of partner superpowers to actually do things on the web under your oversight. MongoDB provides the contextual mission vault, Elastic caches solved UI actions, Arize monitors reasoning and guardrails, Fivetran streams mission analytics, Dynatrace captures runtime telemetry.

How we built it

It was initially built with antigravity and python and javascript. I then moved to my own personal model which led to many iterations.

Challenges we ran into

Many challenges, the main one being access to api's to call llm's and the partner services. these tools are very useful however they all cost and i was limited financially and therefor had limited time with full capability before running out of tokens and credits.

Accomplishments that we're proud of

The concept is solid and the connectivity granted by these partners together provide a strong framework for the app to navigate and perform on a users behalf at rapid pace and efficiency and safely.

What we learned

The more you utilize an agent the better it becomes at a task and that things take longer than anticipated to achieve most of the time.

What's next for Dash Agent

I would like to make it a real thing and if i do get a prize I look forward to engaging with the partners long-term and offering the service as a webapp to the average consumer if possible, i definitely value feedback. Thank you.

Built With

• arize
• dynatrace
• elastic
• fivetran
• gemini
• javascript
• mongodb
• python