[MeDo Hackethon Project] DualSearch,better search

DongQinqin — Sun, 12 Apr 2026 02:49:18 +0000

Inspiration

This project was born from a subtle sense of friction. Whenever I needed to quickly understand a new topic, my screen turned into a battleground: a search engine full of blue links on one side, and an AI chat window waiting for me to paste text on the other. The constant copy-paste-tab switch routine completely sh
attered my train of thought. I kept thinking, Why can't these two steps live in the same place? The idea of merging the "finder of links" and the "giver of answers" into a single input bar planted the seed for DualSearch.

What it does

DualSearch packs two ways of acquiring information into one minimalist interface. Think of it as a search box with a split personality.

Mode One: The Navigator. When you know exactly what website or news article you're looking for, it acts like an efficient librarian. It quickly points you to the right aisle and hands you the exact shelf number (the link).
Mode Two: The Explainer. When you have a vague question like, "Why do I feel sleepy after lunch?", it transforms into a research assistant. It quietly consults the materials, digests the information, and writes out a clear, coherent explanation just for you. At the bottom of the note, it politely lists where the information came from.

Both modes share the same simple input bar, switching with a single click.

How we built it

Behind the scenes, it's a choreographed partnership between two imaginary roles.

First, there's the Speedy Clerk. No matter what you ask, they dash off to the archive room (the internet) and come back with a small stack of the most relevant newspaper clippings and book summaries.

Then, there's the Royal Scribe. If you just want the address (Web Search), the Clerk hands it over directly. But if you want an explanation (AI Search), the Scribe takes the stack of papers from the Clerk, reads through them quickly, and starts writing a summary on a notepad. You can actually see the words appear stroke by stroke (the typing effect). Once finished, the Scribe flips the notepad over and writes down the list of references used.

Challenges we ran into

Training the Scribe presented a tricky issue: Mistaken Identity. Sometimes, the Clerk would bring back books on Botany and Astrophysics, and the Scribe would write a brilliant summary but accidentally credit the astrophysics insight to the botany book. To fix this lack of academic rigor, we had to enforce a strict rule: Every single statement in the summary must be checked against the specific papers the Clerk brought back. No making up sources.

Accomplishments that we're proud of

Seamless Flow: We've made the waiting time feel invisible. By simulating a human typing out the answer, the few seconds of processing time feel like a natural conversation pause rather than a loading screen.
Verifiable Honesty: We successfully ensured that every AI claim is traceable. Users don't just get an answer; they get a path back to the original source. In an age of misinformation, this "auditable trail" is the feature we value most.
Visual Clarity: Despite having two complex engines running under the hood, the user only sees a clean page with a logo and a search bar. There's zero visual clutter.

What we learned

Beyond the mechanics, this project taught us three broader lessons:

Curating beats dumping. Giving the AI three carefully chosen paragraphs is infinitely better than handing it an entire encyclopedia.
Teaching boundaries. Teaching the AI to say "I don't know" is just as important as teaching it to summarize. With clear instructions, it usually stays honest.
Perception is reality. A simple animated cursor can dramatically change how patient a user is willing to be.

What's next for DualSearch

Right now, DualSearch is like an assistant who only browses the public library. Next, we want to give it the key to the user's private study. We plan to connect it to the user's personal bookmarks and browsing history. That way, when you search for a term, it can prioritize by saying: "Actually, you saved a fantastic article about this three months ago—and you even highlighted the key part back then." The goal is to evolve from searching the whole web to searching your personal web.

MeDo - Build full-stack Apps With No-Code AI Platform

Medo: The next-gen platform that gives everyone the power to build

medo.dev

Forem: DongQinqin