Show HN for November 13, 2025

My cofounder and I run a startup in oncology, where we handle cancer genomics data. It occurred to me that, thanks to a recent complexity theory result, there's a clever way to run bioinformatics algorithms using far less RAM. I built this Rust engine for running whole-genome workloads in under 100MB of RAM. Runtime is a little longer as a result - O(TlogT) instead of O(T). But it should enable whole-genome analytics on consumer-grade hardware.

The AI Bubble Monitor is an analytical tool designed to track and visualize indicators of potential market bubbles in AI-related sectors. It aggregates multiple data sources and metrics to produce a composite "AI Bubble Score" that ranges from 0 to 100. The tool breaks down the overall score into five sub-indices: Valuation, Capital Flows, Adoption vs Fundamentals, Sentiment & Hype, and Systemic Risk. Each sub-index provides insight into different aspects of market behavior and potential overvaluation.

Hi HN,

We built an AI agent for data analysts that turns the soul crushing spreadsheet & BI tool grind into a fast, verifiable and joyful experience. Early users reported going from hours to minutes on common real-world data wrangling tasks.

It's much smarter than an Excel copilot: immutable data steps, a DAG of SQL views, and DuckDB for instant crunching over millions of rows. Our early agent prototype ranked #1 on the Spider2-DBT bench. https://spider2-sql.github.io

Try it out and we'd love your feedback!

Thanks, Di Wu & the Shadowfax team

P.S. Shadowfax is Gandalf's horse from LOTR. There's a hidden easter egg site with 3 different triggers, see if you can find them.

Hey HN,

I recently built whatstype.org , a free personality test website that helps people explore their thinking, communication, and relationship patterns.

Unlike most MBTI-style sites that only give you a short label, Whatstype digs deeper:

The test adapts to your responses dynamically

Results are structured around reasoning style, emotional pattern, and social interaction

Each of the 16 personality types includes detailed analysis, strengths, challenges, and real-life advice

No login, no tracking — everything runs client-side

I built it because I was frustrated with the typical “clickbait MBTI quizzes” that don’t respect users’ time or data. Whatstype focuses on clarity, accuracy, and clean UX rather than virality.

Tech stack:

Next.js 14 + TypeScript

TailwindCSS for UI

Structured JSON content for multi-language results

Fully static export, hosted on Cloudflare Pages

If you’re into psychology, data visualization, or just enjoy introspective tools, I’d love your feedback — especially on:

Question design (is it too long / too short?)

Result presentation clarity

Ideas for making the insights more actionable

You can try it here: https://whatstype.org

Thanks for reading!

Generative art has been a long-standing interest of mine. I’ve seen a lot over the years — but this generation was absolutely pure.

I struggle with time blindness and kept missing client Zoom calls. MacOS notification banners appear for 5 seconds and vanish...my brain can't hold onto them. I spent 2 months building a solution.

What it is: Chime shows full-screen alerts that physically interrupt whatever you're doing. It's impossible to miss.

Technical details: - Built entirely in Swift/SwiftUI using Cursor. - Uses EventKit for Calendar + Reminders integration - Todoist REST API v2 for task sync - All data stays local—no cloud sync

How it works: - Syncs from Calendar, Apple Reminders, and Todoist every 1-30 min (This is customisable) - Extracts meeting links from meetings (30+ platforms: Zoom, Teams, Meet, Webex, etc.) - Schedules Timer-based full screen alerts with customizable timing - Sound rotation system to prevent habituation

The insight: Most accessibility focuses on making things "clearer" or "simpler." Neurodivergent brains need "impossible to ignore." Subtle equals invisible. I've been using it for a few weeks and went from missing chunk of meetings to zero.

Try it: 14-day free trial on Mac App Store. No signup required, just download and grant Calendar permissions. https://apps.apple.com/us/app/chime-remember-everything/id67...

Happy to discuss: - Building macOS apps with AI assistance - EventKit's quirks (hourly recurrence workaround, full screen alert implementation, etc) - Designing for neurodivergent users

Hi HN,

I built a simple tool to solve a problem I've faced as a developer and freelancer: how can you definitively prove a file (like code, a contract, or a design) existed at a specific time?

It's called Akashi Notari, and it’s a simple utility for creating an on-chain "Proof of Existence." It lets you anchor any file's cryptographic hash on the blockchain (supports Base, Ethereum, Optimism). And costs less than a dollar.

The main focus is privacy and simplicity:

- *100% Private:* Hashing (SHA-256) is done entirely in your browser. The file never leaves your computer. Only the calculated hash is sent to the blockchain. - *Fast:* The whole process, from file drag-and-drop to a confirmed transaction, takes about 60 seconds. - *Verifiable:* As soon as the transaction confirms, you get a PDF certificate with the hash, timestamp, and a direct link to the on-chain transaction for verification.

I built this as a "poor man's copyright" or a simple notarization tool for:

- *Freelancers:* Proving when a deliverable was completed. - *Creators:* Timestamping IP (art, music, manuscripts) before sharing it. - *Anyone:* Timestamping contracts or records without the cost and hassle of traditional notaries. Maybe some commit-reveal scenarios.

The workflow is just:

1. Drag/drop your file. 2. Connect a wallet. 3. Confirm the transaction. 4. Download your PDF certificate. 5. Prove anytime in the future that your file existed using simple instructions in PDF

You can try it out here: https://akashi-notari.com/ You can also retrieve/verify existing certificates here: [https://akashi-notari.com/certificate-retrieval](https://www.google.com/search?q=https://akashi-notari.com/ce...)

I'd love to hear what the HN community thinks!

I have a couple young kids and they always want to play mad libs. I only really found ones riddled with ads online so I made my own here. I'm a backend guy so go easy on my UI skills. Definitely used Cursor for help on the UI part of it. Anyway hope you play this with your kids too!

Its unpolished v0 so I am releasing it to get general feedback.Built extensively with the aid of AI

I built PolyCouncil to solve a problem I kept running into when testing local LLMs, I never knew if a single model's answer was actually good, or just confidently wrong. Instead of querying one model, PolyCouncil runs multiple local LLMs in parallel through LM Studio, has them evaluate each other's responses using a shared rubric, then conducts weighted voting to reach consensus on the best answer. Think of it like a panel of experts that score each other's work, then vote democratically on the winner. Key features:

Parallel execution across multiple models Rubric-based cross-evaluation (accuracy, clarity, completeness, etc.) Weighted voting system based on peer scores Custom personas (fact-checker, engineer, risk assessor, etc.) Built-in leaderboard to track model performance Single-voter "judge" mode for controlled experiments Clean Python/PySide6 GUI with pre-built Windows executable

Why it's useful:

Compare models systematically without manual evaluation Get more reliable answers through ensemble methods Explore emergent behavior from model deliberation Test how different personas affect model responses

It's free for personal/research use (Polyform Noncommercial). Built for anyone tinkering with local models who wants to see how collective intelligence emerges from LLM orchestration. Would love feedback from folks experimenting with local models! GitHub: https://github.com/TrentPierce/PolyCouncil

I've been building Qantify for the past 2 years - an open-source Python library for algorithmic trading that combines institutional-grade math models with modern ML/AutoML pipelines.

*What makes it different:*

Most trading libraries are either: - Too slow (Backtrader: ~5ms/point, Zipline: ~10ms/point) - Missing advanced math (no quantum finance, chaos theory, optimal execution models) - No AutoML pipeline (QuantConnect has basic ML but no AutoML) - No GPU acceleration

Qantify addresses all of these.

*Performance:* - 0.008ms per data point (50-100x faster than alternatives) - GPU acceleration: 2-3x speedup on compatible hardware - 1767+ comprehensive tests, 100% code coverage - 106K+ lines of code

*Advanced Math Models* (what hedge funds use):on # GARCH volatility modeling from qantify.math.econometrics import GARCHModel garch = GARCHModel(returns) vol_forecast = garch.forecast(horizon=10)

# Heston stochastic volatility from qantify.math.stochastic import HestonProcess, MonteCarloEngine heston = HestonProcess(v0=0.04, theta=0.04, kappa=2.0, sigma=0.3, rho=-0.7) mc = MonteCarloEngine(process=heston) paths = mc.simulate(n_paths=10000, n_steps=252)

# Optimal execution (Almgren-Chriss) from qantify.math.execution import AlmgrenChrissOptimalExecution execution = AlmgrenChrissOptimalExecution( initial_position=10000, time_horizon=timedelta(hours=4), risk_aversion=1e-6 ) optimal_schedule = execution.compute_schedule()

# Advanced cointegration (statistical arbitrage) from qantify.math.stat_arb import JohansenTest, KalmanHedgeRatioEstimator*Unique features* (not found in competitors): - Quantum finance models - Chaos theory applications - Complexity theory - Information theory - Game theory models

*ML/AutoML Pipeline* (end-to-end):thon from qantify.ml import AutoMLRunner, create_features

# Auto-generate 100+ features features = create_features( df, feature_sets=["trend", "momentum", "volatility", "statistical"], windows=(14, 55, 200) )

# Automated model selection with walk-forward validation runner = AutoMLRunner( search="bayesian", # or "grid", "random", "hyperopt" walk_forward=True, # Time-series aware validation test_size=0.25 )

results = runner.run(features, target) # Automatically tests: RF, GBM, XGBoost, LightGBM # Selects best model with proper time-series validation*ML Features:* - Reinforcement learning (DQN, Q-learning, Policy gradients) - Advanced neural networks (LSTM, GRU, Transformers, Attention) - Model drift monitoring - Feature selection (SHAP, mutual information) - Experiment tracking (MLflow, W&B integration)

*Quick example:* from qantify.data import get_candles from qantify.signals import rsi from qantify.backtest.vectorized import run_vectorized

df = get_candles("binance", "BTCUSDT", "1h", limit=1000) df['rsi'] = rsi(df['close'], period=14)

result = run_vectorized( prices=df["close"], long_signal=df["rsi"] < 30, short_signal=df["rsi"] > 70, allocation=0.5 )

print(f"Sharpe: {result.metrics.sharpe:.2f}")*Real benchmarks:* - Backtest speed: < 0.1s (5000 points) vs QuantConnect's ~5-10s - GPU speedup: 2-3x (tested on RTX 3070) - ML accuracy: Market regime 74.3%, Volatility prediction 68.7%

*Why I built it:* I needed advanced math models (GARCH, Heston, optimal execution) combined with modern ML pipelines, all in one library. Existing solutions either lacked the math depth or the ML integration.

*GitHub:* https://github.com/Alradyin/qantify *PyPI:* pip install qantify

I'd love feedback from the HN community. What features would you find most useful? What am I missing?

Built with Python 3.10+, async/await everywhere, modular architecture. Completely open-source and free for personal/research use.

We built an AI tattoo generator + body placement studio. - Generates concepts + previews on real body photos. - Lets users iterate before committing ink. Would love feedback from the HN community on workflow / UX / pricing.