BettaFish

BettaFish is a multi-agent public opinion analysis system that deploys a fleet of five specialized AI agents across 30+ Chinese and international social platforms simultaneously.

You type a topic — say, "Wuhan University reputation" — and BettaFish takes over. A QueryEngine crawls news. A MediaEngine processes short-video content from Douyin and Kuaishou. An InsightEngine mines private databases. A ReportEngine compiles everything into a structured document. And over all of them runs a ForumEngine — a debate moderator that forces agents to challenge each other's findings before anything reaches the final report.

The output is a fully rendered, interactive HTML sentiment analysis report. Not a summary. A document.

Built with zero external frameworks, entirely in Python, under GPL-2.0.

BaiFu (@666ghj) is a 20-year-old senior at USTC (中国科学技术大学). BettaFish started as his university thesis project.

He open-sourced it. It hit 20,000 GitHub stars in the first week.

That performance caught the attention of Chen Tianqiao (陈天桥) — founder of Shanda Group and former richest person in China, now focused on AI and cognitive science. Chen invited BaiFu to Shanda and told him: "Keep doing what you want to do."

While at Shanda, BaiFu spent another 10 days building MiroFish — a swarm intelligence prediction engine. He sent Chen a rough demo video. Chen called back within 24 hours with a ¥30M investment offer.

BaiFu became CEO. He is still in his final year of university.

BettaFish is built on a multi-agent debate architecture — the most technically serious part. The ForumEngine doesn't just aggregate results from other agents; it runs an explicit adversarial review loop, forcing agents to surface contradictions in the data before compiling the final report.

Stack:

• Pure Python — zero framework dependencies
• Five parallel agent types: Query, Media, Insight, Report, Forum
• Multi-modal: processes both text and short-video content (Douyin, Kuaishou)
• Output: fully rendered interactive HTML reports

On the Vibe Coding process (from BaiFu's own writing):

• Most time spent on market research before touching the AI
• Gemini 3 Pro for frontend initialization and UI polish
• Claude for backend architecture and stability hardening
• Multi-agent parallel: run multiple AI agents on the same subtask, pick the best output
• Git discipline: "The faster you move, the more you need good brakes"
• Line-by-line review: "I audit every line the AI writes. I try to understand why it thought this way."

License: GPL-2.0

• Reached GitHub Trending #1 globally after open-sourcing
• Followed 10 days later by MiroFish, also GitHub Trending #1
• Led to ¥30M investment in the follow-on project (MiroFish)
• Coverage in 量子位, multiple major Chinese tech media outlets

Three things set BettaFish apart:

1. Real multi-agent architecture, not a wrapper. Most "AI agents" are a for-loop calling GPT. BettaFish runs five specialized agents in parallel with an explicit adversarial review layer. The ForumEngine that forces debate between agents is a genuine architectural decision, not a marketing claim.

2. Zero framework dependencies. Built in pure Python. This is either a constraint or a flex — in BaiFu's case, it means every part of the system is understood and auditable. He said it himself: "I audit every line the AI writes."

3. The outcome validates the product. 40,000 GitHub stars, 7,500 forks, immediate industry attention, a ¥30M investment in the follow-on project — the market confirmed the thesis before anyone had time to debate it.

超级个体真的能成。 Super individuals can really make it.