Mycellium

Inspiration

The ARDD conference brings together some of the most brilliant minds in longevity science, from researchers to founders to investors, all in the same venue. However, what often fails to occur is the meeting between investor and founder, the connection between researcher and business owner, the coming-together of the community. Conference apps may show users a schedule and a speaker list, but they could do so much more. That's why we built Mycellium: to transform a conference from a static schedule into an interconnected ecosystem.

The name Mycellium comes from the underground fungal networks that connect entire ecosystems together. In the same way, we imagined ARDD as an interconnected network of people, research, ideas, and collaborations waiting to be discovered.

What it does

Mycellium is a full-stack, AI-powered research networking platform specially built for the Aging Research and Drug Discovery (ARDD) conference. It consists of four main modules that work together to curate the conference experience.

Profile Builder & Onboarding A structured five-minute onboarding flow captures not just the basics (name, affiliation, role) but what each attendee is actually there to do: fundraise, collaborate, recruit, publish, or close a BD deal. It also collects research focus, company stage, session interests, and availability across Boston Longevity Week. On completion, attendees are routed into the right Slack workspaces and ARDD-specific channels before the conference even opens.

Personalized Conference Navigator Using the real ARDD programme, Mycellium builds each attendee a custom schedule weighted to their goals — not just their research tags. A founder raising a Series A sees investor panels and company showcases surfaced first. A postdoc there to publish sees dense scientific sessions in their area. The schedule exports as an .ics file that drops straight into Google Calendar.

Matchmaking Engine This is the centrepiece. The engine compares profiles across the full attendee pool and generates specific, grounded match explanations — not "you're both interested in aging" but "your senolytic drug program and their BCL-2 delivery platform address the same therapeutic gap from opposite ends." Match cards display both profiles side by side with shared interests highlighted and the AI reasoning fully visible. The engine supports four core match types: investor ↔ startup, startup ↔ pharma, pharma ↔ pharma, and academic ↔ biotech, plus availability overlap so suggested meetings can actually happen.

ARDD Claw Bot (Slack) A community-level Slack bot serving the entire conference — not just registered app users. Attendees query it with /ardd now, /ardd schedule, /ardd speaker, and /ardd notes to navigate the programme in real time. They submit session impressions via /ardd note, which are aggregated into a shared feed the whole community can query. Organisers push announcements through /ardd announce. At 6 PM each conference day, the bot auto-generates and posts a digest of everything that happened — sessions held, highlights, and notable attendee impressions — and archives the full record for a structured post-conference export.

How we built it

We built Mycellium using a modern full-stack web architecture focused on scalability, real-time interaction, and visualization.

Backend

Node.js
Express
TypeScript
PostgreSQL
Prisma ORM

Frontend

React / Next.js
Tailwind CSS
React Force Graph for the research visualization engine

Real-Time Features

Socket.IO for live messaging and networking interactions

AI / Matching Layer For matchmaking, we combined:

keyword overlap analysis
publication similarity
research-interest embeddings
rule-based explanation generation

Instead of training expensive custom AI models, we used lightweight semantic similarity approaches and structured reasoning systems to create customized recommendations while staying within hackathon constraints.

Infrastructure

To keep the project fully accessible and deployable on free tiers, we used:

Supabase PostgreSQL
Render
Vercel
Cloudinary for video hosting

Challenges we ran into

One of our biggest challenges was creating the Claw Bot as we had never worked with Slack bots before. Among the many issues we ran into, one particularly tricky one was Slack's 3-second ack deadline, which required slash commands to acknowledge an incoming request within ~3 seconds. The /ardd note modal was especially tricky: trigger_id expires ~3 seconds after the slash command fires, so views.open must happen essentially immediately. We solved this by warming a modal view cache at bot startup so the modal open never blocks on a DB query. Thanks to our mentor, Fenil Shah, we were able to solve that issue along with numerous other bugs.

Another major challenge was designing the matchmaking engine in a way that felt genuinely intelligent without relying on expensive large-scale AI infrastructure. We spent significant time thinking about how to make recommendations feel meaningful rather than random. This led us to focus heavily on explainability and relationship reasoning instead of purely numerical similarity scores.

The research graph was also technically challenging because force-directed visualizations can quickly become overwhelming or computationally expensive as the number of nodes increases. We had to carefully structure graph relationships, precompute connections, and optimize filtering to keep the experience interactive and responsive.

We also learned that building a strong networking platform is not just a technical problem — it is a UX problem. Scientific collaboration depends on trust, clarity, and relevance, so we focused on making every interaction feel purposeful rather than overwhelming.

What we learned

This project taught us a lot about:

graph-based relationship modeling
recommendation systems
semantic similarity
scalable backend architecture
real-time systems
scientific UX design

More importantly, we learned how difficult it is to design technology that genuinely helps people connect in meaningful ways. Building Mycellium pushed us to think not just about software engineering, but about how ideas, research, and collaboration spread through scientific communities.

What's next for Mycellium

We see Mycellium evolving into a persistent scientific collaboration platform that extends far beyond a single conference.

Future plans include:

AI-generated research summaries
automated collaboration recommendations
live session intelligence
dynamic community clustering
longitudinal collaboration tracking
integrated paper and citation analysis
conference knowledge archives

Our goal is to help transform scientific conferences from temporary events into continuously evolving networks of discovery. We hope to see you at ARDD 2026 in Boston!