IdeaRadar

IdeaRadar helps founders compare a startup idea against a real market corpus. It is not a chatbot. It is a search-and-compare app that uses Qdrant to surface similar companies, explain overlap, and suggest a sharper wedge.

What it does

• Paste an idea into the text box.
• Upload a document: DOCX, TXT, MD, Markdown, JSON, or CSV.
• Paste a live project URL and extract the page title, description, and visible text.
• Search a real company corpus indexed in Qdrant Cloud.
• See the closest matches, why they overlap, what differs, and how to position the idea better.
• Review a semantic landscape, saturation meter, and top-three comparison panel.

Architecture

flowchart LR
  A["Founder input"] --> B{"Source mode"}
  B --> C["Text"]
  B --> D["Doc upload"]
  B --> E["Live project URL"]

  D --> F["Extract text on server"]
  E --> G["Fetch title, description, page summary"]
  C --> H["Idea text"]
  F --> H
  G --> H

  H --> I["Hybrid search request"]
  I --> J["Qdrant Cloud"]
  J --> K["Dense vector search"]
  J --> L["Sparse keyword search"]
  K --> M["Fusion + ranking"]
  L --> M
  M --> N["Result analysis"]
  N --> O["Overlap chips"]
  N --> P["Differentiation wedge"]
  N --> Q["Semantic landscape"]

Loading

Search flow

1. The user chooses one source mode.
2. The app converts that source into normalized idea text.
3. The search API sends the text to Qdrant Cloud.
4. Qdrant retrieves matches using dense + sparse hybrid search.
5. The app runs a lightweight analysis pass to explain:
   • why the match is similar
   • what is different
   • what wedge to use

Data pipeline

flowchart LR
  A["YC companies JSON"] --> B["Ingestion script"]
  B --> C["Build dense text"]
  B --> D["Build BM25 keywords"]
  B --> E["Optional logo embedding"]
  C --> F["Qdrant dense vector"]
  D --> G["Qdrant sparse vector"]
  E --> H["Qdrant visual vector"]
  F --> I["Qdrant collection"]
  G --> I
  H --> I

Loading

Stack

• Next.js 16
• TypeScript
• Qdrant Cloud
• Qdrant Cloud Inference for dense embeddings
• BM25 sparse retrieval
• Local CLIP embedding for optional logo/vector support

Source modes

Choose one way to search at a time:

• Text
• Doc upload
• Live project URLs

The inactive inputs are muted so the flow stays clean.

Setup

1. Install dependencies.
2. Copy .env.example to .env.local.
3. Add your Qdrant Cloud values.
4. Start the app:

npm run dev

Corpus ingestion

The corpus is a YC company dataset stored offline in data/yc-companies-all.json.

To index it into Qdrant Cloud:

npm run ingest:yc

Environment

Minimum required environment values:

QDRANT_URL=https://your-qdrant-cloud-cluster-url:6333
QDRANT_API_KEY=your_qdrant_cloud_key
QDRANT_COLLECTION=idea_radar_yc_companies
QDRANT_INFERENCE_MODEL=sentence-transformers/all-MiniLM-L6-v2
QDRANT_SPARSE_MODEL=qdrant/bm25

Demo flow

1. Open the deployed app.
2. Pick a source mode.
3. Paste the idea, upload a doc, or paste a live project URL.
4. Click Find Similar Companies.
5. Review the result cards, semantic landscape, saturation meter, and comparison panel.

Future plan

• Voice-to-text input for idea capture.
• Better comparison and wedge explanations.
• Expanded source support if needed later.