[verdict][curated prompt][source: ideafast.pro]
Can I vibecode IdeaFast?
// buildable in a weekend, but real gaps stay open
The pipeline is honest work an agent can do: pull public Reddit JSON, prefilter complaint-shaped text, classify with an LLM, embed and cluster, rank by frequency times severity times recency. A weekend gets you ranked pain themes with real permalinks for two or three subreddits you already know. What does not fall out of one session is everything after the demo: staying inside Reddit's rate limits at scale, picking which communities are worth scanning when you do not already know, deduping the same pain across runs so week two is not week one again, and keeping the LLM bill under the price of the subscription. Verdict is kinda, not yes, because the first run is easy and the tenth is where the product actually lives.
confidence: medium
what it is: Scans Reddit conversations, clusters them into scored pain themes backed by real quotes, and turns the strongest into startup ideas
Buildability index · an editorial game
- Price 19 $/month weight: plus 2
- Time multi-day weight: plus 1
- Category user research no weight
- Moat infrastructure scale · execution quality weight: minus 3
- Confidence medium weight: zero
- What you lose 5 items weight: minus 2
- Site ideafast.pro (si apre in una nuova scheda) no weight
The moat outweighs the price: rebuilding this is a project, not an evening.
Gioco editoriale: il verdetto dice se un agente può, l’indice se conviene.
What you build
Pull posts and comments from a handful of public subreddits, classify complaints with an LLM, cluster them into named pain themes, and rank them with clickable quote evidence.
what you need
- Anthropic or OpenAI API key
- embedding model
- SQLite
- patience with Reddit rate limits
Good kinda entry: the demo is a weekend, the product is the ingestion plumbing behind it.
The prompt
A weekend with a coding agent. The gaps that stay are right below, under “what you lose”.
Build me a Reddit pain finder: a local CLI plus a small dashboard that reads public
Reddit and turns complaints into ranked pain themes with clickable evidence.
- TypeScript on Node 22, SQLite via better-sqlite3, Hono for the dashboard. One
repo, one `npm run scan` entrypoint. No accounts, no cloud, no telemetry.
- Input: subreddits.txt plus a timeframe flag (default 90 days). Fetch posts and
top level comments from Reddit's public JSON endpoints, for example
https://www.reddit.com/r/<sub>/top.json?t=year. One request every 2 seconds, a
real descriptive User-Agent, and cache every raw response in SQLite so re-runs
cost nothing. No OAuth, no logged-in scraping.
- Prefilter to complaint-shaped text with cheap regexes ("I hate", "why is there
no", "wasted hours", "workaround", "gave up on") before spending a single
token. This is the whole cost story, do it first.
- Classify survivors with Claude (ANTHROPIC_API_KEY in .env, batched, cached by
content hash) into: is_pain, severity 1 to 5, one line summary.
- Cluster: embed the summaries, group at cosine similarity above 0.82, then have
Claude name each cluster and pick its 5 strongest verbatim quotes with
permalinks. Never paraphrase a quote, evidence has to be clickable or it is
worthless.
- Score each cluster as frequency x mean severity x recency decay, and persist
scores per run so a later scan can show what moved.
- Dashboard on localhost:3000: ranked clusters, expandable quotes with permalinks,
filter by subreddit, CSV export.
- README: how to choose subreddits, the rate limit rule and why breaking it gets
you blocked, and rough token cost per 1000 comments.
- Out of scope: sources other than Reddit, idea generation, and cross-scan dedupe.
Get one subreddit list producing clusters you actually trust first.Curated prompt: written and reviewed by hand for this app. In English on purpose — it is the language coding agents work best in.
What you lose
- community discovery, you can only scan subreddits you already thought of
- cross-scan dedupe, so repeat runs resurface the same pains as if they were new
- a warmed corpus, every fresh scan pays the full ingestion wait
- cost control, naive LLM classification of a busy subreddit gets expensive fast
- the idea generation and validation layer on top of the raw clusters
Why people still pay
The clustering is not the hard part, the boring infrastructure around it is. Reddit throttles aggressive clients, so a real corpus takes patient background ingestion rather than a scan you kick off and watch. People pay to skip the warm-up and the API bill, not because the data is secret. It is all public.
moat: Infrastructure scale Execution quality what a moat is
ingestion pipeline + cost control, not data ownership
Who has already built it
Starting from here is still vibecoding: the prompt is for when you want it exactly your way.
- PRAW (opens in a new tab) — Python Reddit API wrapper, the usual starting point for the ingestion half (alive)
- BERTopic (opens in a new tab) — Topic clustering over embeddings, covers the grouping step without an LLM (alive)
Do you agree?
The vote balance
Ancora nessun voto: il tuo è il primo.
Nessun voto ancora — il primo pesa.
