Rodney, your ResearchBuddy

Inspiration

We're motivated by intrinsic human curiosity, particularly in people managing medical conditions and in marginalized communities. Our mission is to make scientific knowledge more equitable and understandable.

What it does

Rodney is on a mission to democratize access to science and medical research to curb the prevalence of medical misinformation.

How we built it

Rodney is a web app built with React and Tailwind, backed by DigitalOcean's Gradient AI Platform. A DigitalOcean GenAI agent (running Claude) powers the core reading experience — plain-language summaries, highlight-to-explain, and adjustable reading levels, all grounded in the actual paper's text rather than a generic AI answer. We also built real multimodal generation on top of DigitalOcean's serverless inference: Flux for illustration generation and ElevenLabs for narration, orchestrated into a "Watch mode" that turns a paper into an auto-narrated visual walkthrough. Lightweight personalization (competency level, topic interest) tailors both explanations and a "picked for you" reading list pulled from arXiv's own API. The whole thing is deployed live on DigitalOcean's App Platform.

Challenges we ran into

Working solo with no prior coding background meant learning Git, deployment, and backend concepts in real time under a hard deadline. DigitalOcean's async image/audio generation API added real friction — status responses didn't match what the documentation implied (COMPLETED vs. the expected COMPLETE), which took direct testing against the live API to catch. We also had to scope hard: true video generation would've required standing up GPU infrastructure we didn't have hours for, so we made the deliberate choice to build genuinely real image and audio generation via DigitalOcean, and cache those results ahead of time rather than risk a live generation hang during the demo.

Accomplishments that we're proud of

Shipping a working, deployed, end-to-end multimodal AI product — real RAG-backed explanations, real AI-generated imagery, real AI-generated narration — solo, from a standing start with no coding background. We're especially proud that every "AI feature" in this build is genuinely wired up rather than mocked, and that we made honest tradeoffs (like caching demo assets instead of faking a broken live generation) rather than over-promising what wasn't ready.

What we learned

How retrieval-augmented generation actually works in practice, not just in theory. How to design around the real constraints of a live demo — knowing when to fake something responsibly (a slideshow instead of true video) versus when it's worth the risk to build the real thing (image and audio generation, which we did get working on real infrastructure). And a lot about scoping: cutting features aggressively so what remains actually works beats trying to ship everything and having none of it be reliable.

What's next for Rodney, your ResearchBuddy

A browser extension so Rodney recognizes a paper the moment you're reading it, with no copy-pasting a link required. True video generation via a GPU-backed model once we have the infrastructure to support it reliably. Real accounts and a persistent reading dashboard across devices, beyond the local, single-browser version we built for the demo. And deeper personalization — using a reader's history with Rodney to get better at meeting them at their level over time, not just from a one-time onboarding question.

Built With

• rag