Partliament

Inspiration

All 3 members of this team have some high school robotics experience. One of the worst, most logistically taxing tasks we've all had to complete is making a BOM (Bill of Materials)—an itemized list of parts that go into making our finished product. The process involves a lot of complex human deliberation: detailed cost-benefit comparisons and accurate assessments of resource availability. Partliament brings the discussion of an engineering team to a single interface: multi-agent specialists research, challenge each other, and negotiate a parts list around your mission and budget.

What it does

Users describe a device in plain language, such as an underwater rover that collects temperature data, and set a budget. This sets a variety of agents into motion. A facilitator identifies the relevant engineering disciplines. Field experts define requirements, and parts experts research products across supported vendors. Three final reviewers, focused on cost, electronics, and hardware, negotiate a bill of materials. Users can watch the discussion and inspect product links, quantities, price estimates, and cost breakdowns. A final BOM requires unanimous approval of the same proposal and successful validation of quantities, requirement coverage, and budget.

How we built it

We built a Python FastAPI backend with custom asynchronous agent orchestration and a lightweight HTML, CSS, and JavaScript frontend. The backend supports OpenAI, NVIDIA NIM, and ElevenLabs model integrations. Optional AGNTCY integration carries agent messages through NATS.

Challenges we ran into

AI-generated designs looked generic. We tried to iterate to give Partliament a clearer visual identity.
Multiple agents researching and negotiating meant many API calls.
Some models were better at answering individual questions than responding meaningfully to other agents. We needed clearer roles and structured discussion rules to make negotiation productive.
Adopting AGNTCY, an unfamiliar SDK meant understanding its ins and outs: its messaging abstractions, NATS, and connecting message delivery to our existing workflow.

Accomplishments that we're proud of

We’re particularly proud of our ability to make disagreements between individual agents visible. The system can come to complex decisions, report that it has not reached agreement, and defer. This was also two of our members' first hackathon! We're very proud of having a complete finished product.

What we learned

We learned that getting multiple agents to work together is difficult. Each agent needs a clear role and a way to resolve disagreements. We also learned that even when every agent agrees, the design still needs to be checked by someone who understands the hardware.

What's next for Partliament

We plan to support larger projects with durable background checks on materials vendors. We also hope to help users move from a proposed BOM toward a validated build.