Tulkas

Inspiration

Two things inspired me: the first was a recent news cycle about evil at the highest levels of power, and the second was the Tolkien-named companies taking a large part in related evil. I wanted to build a Tolkien-named tool that helped the world become a more ethical place, a place more in line with his values. Tulkas is named for one of only two gods who saw through Melkor-Morgoth's lies.

What it does

Tulkas operates on three levels: the database level, the basic search level, and the advanced search level. A cache stores recently searched entities for efficiency. The database level stores several prominent figures and organizations with set scores. The basic search level uses a capped AI, sourced on DO Gradient, and Brave Search to do a short investigation and return a score. The advanced searched level uses AI agents, no cap on resources, and user input to give a more fulsome and researched score and perspective. All levels use sources that the user can personally validate.

How we built it

Tulkas was built with heavy use of AI. I had it create, and personally checked, a six-stage plan. It then built the app out, and I rigorously tested each stage.

Challenges we ran into

There were three major challenges. First was the provisioning and use of all the resources; I had never used Valkey before, for example, nor any AI agents. I had to figure out how to interact with many new features and services on DO before I could do any coding, and that took some time. Second was the AI workflow and use. There were several prompt and model errors surrounding the amount and interpretation of research by the LLM; it took many iterations to get even basic search working, and then more to figure out the integration of the advanced search agent. Third, and still unsolved, was the minor issue of marketplaces. I had wanted to have Tulkas score products on Amazon's product pages and score Amazon itself on the homepage, but my solutions for this ended up breaking many aspects of the app; I'm still working on them, but Tulkas is usable for many media sites such as Wikipedia, X/Twitter, Instagram, etc.

Accomplishments that we're proud of

I'm proud I was able to complete this project, use AI agents for the first time, maintain the vision while the AI generated the code.

What we learned

I learned a lot about AI integration and the possibilities of agents.

What's next for Tulkas

That marketplaces issue, and trying to get the extension onto the Firefox/Google stores

Built With

• brave-search-api
• chrome-extensions-api
• digitalocean-app-platform
• digitalocean-managed-postgresql
• digitalocean-managed-valkey
• digitalocean-spaces
• express.js
• gradient-ai-adk-agent
• gradient-ai-inference-api
• gradient-ai-knowledge-base
• node.js
• preact
• typescript
• vite
• webextensions