Jac of all Trades

Inspiration

The kernel of this idea began with Derek. His research scientist and builder mindset latched onto the idea of graph based architectures and abstracted LLM calls after hearing about the Jaseci and reading the "MTP: Meaning Typed Programming" whitepaper. He called me (Luke) & our teammate Justin up with the idea that we could build a financial platform that allowed people like our parents (and even ourselves) to make more informed decision about finances, and that we should explore this idea at JacHacks. Our teammate Amogh noticed Derek's propensity to think outside the box during the opening ceremony and after talking with him for a while, we decided to team up. Thus, Jack of all Trades was formed.

What it does

Jack of all Trades is an ACP agent that interacts with your portfolio via API/MCP using natural language to determine what portfolio data is relevant to the users question, then it uses another LLM to generate an appropriate visual representations of the synthesized data.

How we built it

We built the project by creating an ACP agent, working with Alpaca API/MCP, and a Jac Python/JS frontend environment. Also caffeine, and panda express, a littlel bit of grit, but mostly caffeine.

Challenges we ran into

We ran into a number of challenges working with the Jac tooling, it was very new to all of us so the learning curve of all the nuances was high. These problems ranged from the inability to add a Claude API key to JacCoder, JacCoder breaking and being unable to relaunch, and the limited training data that exists on the language. This being said, a majority of these problems could be mitigated / resolved in the future given more time and skill utilizing the toolset, but for a 24-hour time window, this was our most challenging hurdle.

Accomplishments that we're proud of

Collaborating with new people, building a makeshift bed in the 4th floor grad student lounge, pushing through the challenges to have a great experience and getting a demo made.

What we learned

A lot. First that NVIDIA, Anthropic, and OpenAI are cooked. Second, Jaseci is a really cool technology developed by a team who has given their blood, sweat, and tears towards finding a better way to be computing in the era of AI. Finally, that the technology is still raw, needs work, needs testers, and needs applications built on it to understand what it's true potential really is.

On a more pragmatic note, we felt like we got a much deeper understanding of Agents, MCP Servers, and the entire agentic ecosystem.

What's next for Jac of all Trades

There was a moment around 1 AM where Derek went on a 3-4 Zip bender and theorized the idea that our users financial portfolio could be represented in Jaseci as a graph, then an agent could traverse this graph to collect the relevant information for the users query, really exciting part of this idea is that the walker mechanism could produce an artifact based on how this data is connected in ways we normally wouldn't analyze, and that artifact could be utilized as a new input for financial analysis. I think we are all wondering what kind of insights that walk could provide as it is such a unique piece of information, and personally I want to explore this idea to see if it has any merit.