How IdeaCheck Works

Step 1 — Embed your idea

When you submit an idea, we convert the text into a 768-dimensional vector using BAAI/bge-base-en-v1.5, a state-of-the-art open-source embedding model. This vector captures the semantic meaning of your idea rather than just keywords.

Step 2 — Retrieve similar Hacker News discussions

We run a cosine similarity search over our pgvector index of 28,000+ Hacker News threads — mostly Ask HN: Show Your App, Show HN, and founder AMAs. The top matches are threads where the HN community already discussed ideas similar to yours: same market, same problem, or same technical approach.

Step 3 — Attach community reactions

For each retrieved thread we attach the highest-voted comments. These comments carry the actual community signal — skepticism, enthusiasm, alternative solutions, or "this was tried by X and failed for Y reason." The LLM uses this context to ground its verdict in real founder discourse rather than generic startup advice.

Step 4 — Score with an LLM

The retrieved context plus your idea are sent to Llama 3.3 70B (via Groq) with a structured prompt that requests four scores (0–10) and a final verdict:

• Market demand — Is there evidence people want this?
• Technical complexity — How hard to build and maintain?
• Distribution — How easy to reach first customers?
• Timing — Is the market ready now?

The LLM returns a JSON blob with scores, a one-paragraph rationale per dimension, and a final verdict of PASS, PROMISING, or PASS ON. The scores stream to your browser in real time.

What IdeaCheck is not

IdeaCheck is a fast first-pass filter powered by community signal, not a market research firm. A PASS verdict does not guarantee success. A PASS ON verdict does not mean your idea is bad — it means the Hacker News community was skeptical of similar ideas, which is worth knowing before you spend months building.