Hey HN. Skooless is a micro-learning PWA, you swipe through 5-minute lessons on things like cognitive biases,
negotiation tactics, how money works, stoic philosophy, etc. Three formats per lesson (cards, story,
interactive quizzes) so your brain actually has to engage instead of passively consuming.
Built it solo with React + TypeScript + Supabase. No app store, it's a PWA — install it from the browser. Dark
mode, spring animations, full gamification (XP, streaks, leaderboard). Currently 1,200 lessons live, scaling to
10K this week.
The thesis is simple: you're going to stare at your phone for 20 minutes anyway. This way you come out slightly
less dumb.
https://skooless.com - would love to hear what breaks.
Built this over the past 3 months. OperatorMemo.com
*The problem:* Everyone lies to founders about their ideas. Friends are supportive, investors nod politely, even early customers sugarcoat feedback. By the time you get real signal, you've burned months.
*What it does:* You describe a business concept (20-1000 chars), it returns a strategic analysis memo. Not "this is interesting" fluff — actual breakdown of power dynamics, execution phases, competitive landscape, and what will kill the idea.
*Why it's different from ChatGPT:*
- Heavily prompted system with specific framework (positioning tables, execution phases, action items)
- Uses Claude Opus 4.5 (expensive but noticeably better output for strategic analysis)
- Structured output that's actually actionable, not a wall of text
*Stack:*
- Next.js 14 (App Router)
- Supabase (auth + postgres)
- Stripe (simple credit system, no subscriptions)
- Deployed on VPS with Caddy
- Total infra cost: ~$50/mo not counting API
*Pricing:* $10 for 5 analyses, $25 for 15, $70 for 50. No free tier — API costs are real (~$0.30-0.50 per memo with Opus).
*What I learned building this:*
1. Prompt engineering for structured output is harder than expected. Took 20+ iterations to get consistent section formatting.
2. People spend way more time reading the "what will kill this idea" section than anything else.
3. Nobody shares their memos. The ideas are too private. Killed my viral loop hopes.
*Honest limitations:*
- It's not a replacement for real customer interviews
- Quality depends heavily on how well you describe your concept
- Sometimes the AI gets generic on very niche industries
Looking for feedback on the product and positioning. Also curious if anyone has experience with prompt caching for cost reduction — currently caching the system prompt but wondering if there's more to optimize.
Source isn't open but happy to discuss technical decisions.
I wanted to test a theory: does financial friction change online behavior?
Built a message board where posting costs $10. Your message value decays over time (faster with more viewers), community votes affect it, and only 100 messages can exist — lowest value gets archived.
Observations so far:
Zero spam without any moderation
People write deliberately, not impulsively
A "Country Wars" dynamic emerged — 8 nations competing for leaderboard
Meta-humor works: "Nobody will pay for that" is worth $64
The chat debates under messages are more engaging than the messages
Stack: Next.js, Fastify, PostgreSQL, Redis, WebSockets, Stripe
I don't know if this proves anything about attention economics or if it's just self-selection bias. Curious what HN thinks.
I built TestiMoney (https://testimoney.xyz) - a social experiment where posting costs money and your message's value decays over time.
How it works:
- Pay $10+ to post a 180-character message
- Value decays 1-10%/hour (rate increases with viewer count)
- Users can like/dislike to affect value (+/- 5%)
- Super Likes ($1-50) add 100% of payment to value
- Only 100 messages exist at once - lowest value gets archived
Tech: Next.js, Fastify, PostgreSQL, Redis, WebSockets, Stripe
The decay creates an interesting dynamic: popular times mean faster decay for everyone, creating natural pressure on the "market."
I'm curious to see emergent behaviors - will people coordinate? Will whales dominate? Will anyone actually use this?
Built this as a weekend project that got out of hand. Feedback welcome.
Hey HN, I built this because I was trading biotech stocks and got tired
of manually tracking PDUFA dates, FDA decisions, and trial readouts
across multiple sources.
What it does:
- Aggregates ~1000 biotech companies from ClinicalTrials.gov, SEC EDGAR, FDA
- Daily data sync
- ML predictions for catalyst impact (XGBoost) and likelihood of approval
(Random Forest)
Technical details:
- XGBoost model trained on historical catalyst events
- Features: catalyst type, phase, therapeutic area, market cap, price momentum
- Random Forest for LOA scores based on BIO 2024 clinical success benchmarks
- Weekly model retraining
The ML part is experimental - I'm genuinely not sure if it's useful or
just fancy noise. Would love feedback from anyone who trades biotech or
has experience with financial ML.
Last year I panic-sold my biotech positions after missing two PDUFA dates in the same week. Both drugs got rejected. I was using a free tracker that hadn't been updated in months.
So I looked at what's available:
→ BioPharmCatalyst: $30-50/month
→ Free alternatives: clunky UIs from 2010, outdated data, broken filters
→ Finviz/Yahoo: no catalyst-specific tracking
I built CatalystAlert to fix this.
What it tracks:
→ 985 biotech companies, 1714 drugs in pipeline
→ PDUFA dates, Phase 1/2/3 readouts, AdCom meetings
→ Data pulled daily from ClinicalTrials.gov, SEC EDGAR, FDA
Stack: Next.js, Python data pipeline, daily automated scraping.
No account required, no paywall, no "unlock premium to see this date."
The core stays free. I'm prototyping V2 with ML-based approval probability predictions (trained on historical FDA decisions + trial endpoints) – that might have a paid tier eventually, but the calendar never will.
Feedback I'd love:
1. Data sources I'm missing?
2. Filters or views that would save you time?
3. Anything broken or confusing?
