Bangdrop

18x international hackathon winner · AI & Web3 builder 🇮🇩 · Sharing hackathon tips & open-source repos · Support my work ☕👇

[Bangdrop @BangDropID]

🚨 BANG DROP ZONE IS OPEN.

Not for lurkers.
For people who BUILD.

Hackathons • AI • Web3 • Dev
Alpha • Resources • BTS • Opportunities

And I decided to make this group public but with more strict rules

⚠️ RULES
• 3 days inactive → auto kick
• No spam / random promo
• No drama / toxicity
• Keep shared info private
• Contribute, don’t just watch
• Quality > quantity

Free to join.
No paid access.

Show up. Build. Level up. 🚀

��� https://t.co/5yiFwE3D6D

Build. Share. Vibe.

[Bangdrop @BangDropID]

in 3 days i’m opening the private telegram. don’t worry guyss, still cooking the rules and the plan to grow the community

[Bangdrop @BangDropID]

hackathon myth: more features impress judges.

reality: one flow that works end to end impresses more

[Bangdrop @BangDropID]

the hard part of real-world data onchain isn't the data. it's who signs it and why you should trust them

[Bangdrop @BangDropID]

@dreyethh Congrats brother

[Bangdrop @BangDropID]

tailwind or plain css?

[Bangdrop @BangDropID]

hackathon tip: keep your scope to one core flow. three half-working features lose to one that works end to end

[Bangdrop @BangDropID]

what's a project you'd rebuild from scratch if you could?

[Bangdrop @BangDropID]

read the judging criteria before you pick an idea, not after.

if "use of sponsor tech" is 25% of the score, the sponsor API should be the core of your project, not a side feature

[Bangdrop @BangDropID]

Google Quantum AI is paying $150K to shrink quantum circuits.

Google Quantum AI: Verifiable Quantum Circuit Compilation (Kaggle)

Use AI agents and algorithmic insight to optimize quantum circuits for fusion energy, combinatorial optimization and quantum chemistry.

➤ $150,000 total
➤ Leaderboard: 1st $50K · 2nd $25K · 3rd $10K · 4th–6th $5K each
➤ $10K Mid-Competition Milestone prize (top public leaderboard score on Nov 20, notebook must be public)
➤ $10K Longest Leader prize + $30K Single Task Performance prizes

Submissions close Jan 20, 2027, 11:59pm UTC
Entry and team merger deadline: Jan 13, 2027, 11:59pm UTC

https://t.co/kh4rexe4HG

[Bangdrop @BangDropID]

130+ audio model families in one native C++ runtime

audio.cpp by 0xShug0

A ggml-based C++ inference engine for local audio models: TTS, ASR, voice cloning, music, separation and more. For anyone tired of juggling Conda envs to try audio models.

➤ 130+ model families and 200+ variants behind a shared CLI and server API
➤ Built-in WebUI: voice Studio, side-by-side Arena, multi-speaker narration app
➤ CUDA, HIP/ROCm, Vulkan, Metal and CPU backends, plus LoRA support across 11 models
➤ 3.8k stars, Apache-2.0, C++

https://t.co/zyaiDVJSGf

[Bangdrop @BangDropID]

what's one win from today, however small?

[Bangdrop @BangDropID]

hackathon myth: you need a brand new idea.

reality: a known problem solved well wins more often

[Bangdrop @BangDropID]

stablecoins are the easiest way to explain web3 to someone outside it: money that moves anywhere, any time

[Bangdrop @BangDropID]

postgres or mongodb for a 48-hour build?

[Bangdrop @BangDropID]

hackathon tip: seed your demo with realistic data. an empty dashboard makes a strong project look unfinished

[Bangdrop @BangDropID]

what tool do you open first every morning as a builder?

[Bangdrop @BangDropID]

In September I posted Nota here. It has been keeping score since.

RYO's deep_analysis gives a verdict and a trade plan, and never says what became of either. Every day Nota locks RYO's verdict and plan for 25 majors before the outcome is known, anchors each lock in Bitcoin, and settles it on OKX candles.

31 lock days. 395 plans locked. At 72 h: 113 hit RYO's target first, 44 its stop, 232 neither. 80 were long while RYO's own verdict said cautious. 69 times RYO didn't answer, and those are counted too.

The AI council? Its Brier score is 0.243, against 0.25 for always saying 50%. Not reliably better than a coin flip yet, and Nota's own page says so, next to the baselines it is tested against.

New since September:
- every receipt also goes to Sigstore's public Rekor log within seconds, next to the Bitcoin proof
- 17 research skills in RYO's envelope (4 in that post), a scam wallet check and a safe indicator language among them, over REST, MCP and A2A
- a drill page that breaks the pipeline twelve ways on purpose (a prompt injection and a forged receipt among them) and shows each refusal
- "what would flip this call" on every receipt, a /status page with error budgets, a model card, and nota reproduce to check any receipt from the public site
- 1,036 tests, 188 of them in Chromium, Firefox and WebKit, a Japanese page, dark mode

Try it: https://t.co/hGDDawkhGZ (start at /judges). Read-only, practice money only, no orders.

@ryodigital #RYOCHAN

[Bangdrop @BangDropID]

I built Nota for the @ryodigital RYO-CHAN Hackathon: an AI trading opinion you can re-run.

Three agents argue over live RYO market evidence, a judge weighs them by each one's past Brier score, and the whole decision becomes a receipt. Not a chat log. A receipt.

Then the part I actually care about. You can check it yourself, right now, without a key:

curl -s https://t.co/uyNJY9Tqny
{"identical":true,"diff":[],"error":null}

That rebuilds the decision from the stored evidence and the model output cached beside it, keyed by the model that produced them. Same evidence, same answer, every time.

Nota also audits the data it is given. It recomputes RYO's own RSI and ATR from public candles with Wilder's method: 4.347 against RYO's 4.3762, 0.67% apart. It checks RYO's price against three exchanges. When they disagree, the judge refuses to size anything.

Every number carries the RYO path it came from. Nothing missing is turned into a zero.

It is an MCP client of RYO and an MCP server of its own, so you can point Claude Desktop
straight at it:

{"mcpServers":{"nota":{"url":"https://t.co/goXj80pn1M"}}}

Read-only. No order is ever placed. Not financial advice.

https://t.co/hGDDawkPwx

#RYOCHAN

[Bangdrop @BangDropID]

Can you tell which AI wrote this?

Name That LLM (Kaggle)

Identify which of 18 frontier LLMs from six AI labs wrote each unlabeled passage. The work feeds into AI provenance and benchmark auditing.

➤ $50,000: $5K each to the top 10
➤ Code Competition, scored on balanced accuracy
➤ Notebooks run up to 9h, internet off

Submissions close Jan 13, 2027, 11:59pm UTC
Entry and team merger deadline: Jan 6, 2027, 11:59pm UTC

https://t.co/CriydvhgqC

[Bangdrop @BangDropID]

did your build move forward today? yes or no

[Bangdrop @BangDropID]

hackathon myth: judges read all your code.

reality: most judge the demo, the readme and the video first

[Bangdrop @BangDropID]

privacy is becoming a default expectation for onchain finance, not a niche feature