MarketShift

Competitive intelligence from a single product URL. Paste a competitor’s or your own product URL; MarketShift discovers similar companies, gathers news (launches, funding, partnerships, market moves), classifies signals, and surfaces market opportunities, competitive risks, and suggested next steps. Optional: get a short spoken briefing (audio summary) of the results.

---

What it does

• One input: A product or company URL (e.g. Notion, Coda, your own landing page).
• Pipeline: Scrapes the URL, finds the top 5 relevant competitors (Exa + LLM), then for each competitor and the market runs news, web, and Product Hunt searches.
• Structured output: An LLM classifies items into product announcements, funding, partnerships, and market signals, then synthesises market opportunities, competitive risks, and action steps.
• Dashboard: A single view of “what changed” and “what to do” (product announcements, funding, partnerships, market signals, opportunities, risks, next steps).
• Voice brief (optional): Generate an MP3 summary (30–60 seconds) of the analysis via text-to-speech.

---

Tech stack

Layer	Stack
Frontend	React 19, Vite, Tailwind CSS v4 (JavaScript)
Backend	Python, FastAPI
Search	Exa (company/news search, scraping)
LLM	DeepSeek (V3/R1) via GMI Cloud — intake, analysis, narrative summary
TTS	ElevenLabs via GMI Cloud (voice summary)

---

Quick start

1. Backend

cd backend
source venv/bin/activate
pip install -r requirements.txt

Set environment variables (see Environment), then:

uvicorn main:app --reload

API: http://localhost:8000

2. Frontend

cd frontend
npm install
npm run dev

App: http://localhost:5173

Frontend proxies /api/* to the backend (see frontend/vite.config.js).

3. Run the pipeline

• POST a product URL to run the full pipeline:

  curl -X POST http://localhost:8000/pipeline \
    -H "Content-Type: application/json" \
    -d '{"url": "https://www.notion.so"}'

• POST voice summary (uses latest pipeline results): POST http://localhost:8000/pipeline/voice-summary → returns MP3.

---

Environment

Create a .env in backend/ with:

Variable	Purpose
EXA_API_KEY	Exa search and content API
GMI_API_KEY	GMI Cloud (DeepSeek LLM + ElevenLabs TTS)

---

Project layout

MarketShift/
├── frontend/                 # React + Vite + Tailwind
│   ├── src/
│   │   ├── App.jsx
│   │   ├── LandingPage.jsx   # Hero input → AgentTimeline → Dashboard
│   │   ├── AgentTimeline.jsx # Research progress UI
│   │   └── MarketShiftDashboard.jsx  # Signals, sentiment, competitors, actions
│   └── vite.config.js        # API proxy /api → :8000
├── backend/
│   ├── main.py               # FastAPI app, /pipeline, /pipeline/voice-summary
│   ├── agents/
│   │   ├── orchestrator.py   # Pipeline: intake → sub_1 → sub_2 → analysis → sub_3
│   │   ├── intake_agent.py   # URL → scrape → competitor discovery (Exa + LLM)
│   │   ├── sub_agent_1.py    # Per-competitor & market news/web/Product Hunt
│   │   ├── sub_agent_2.py    # Aggregate, dedupe, persist to data/results.json
│   │   ├── analysis_agent.py # LLM classification + opportunities/risks/actions
│   │   ├── sub_agent_3.py    # Shape output for frontend
│   │   └── voice_summary.py  # Pipeline output → narrative → TTS → MP3
│   └── data/                 # results.json, voice_summary.mp3 (gitignored)
└── AGENTS.md                 # Developer/agent notes (commands, architecture)

---

API overview

Method	Path	Description
POST	/pipeline	Full pipeline: single product url → structured intelligence
POST	/pipeline/voice-summary	Generate MP3 briefing from latest pipeline results

---

Pipeline data flow

1. Intake — Scrape product URL (Exa), find similar companies (Exa), LLM picks top 5 competitors and market context.
2. Sub-agent 1 — For each competitor and the market: news, web, Product Hunt searches (Exa).
3. Sub-agent 2 — Deduplicate by URL, tag source, append run to data/results.json.
4. Analysis — LLM classifies items into product announcements, funding, partnerships, market signals; synthesises opportunities, risks, action steps.
5. Sub-agent 3 — Format for frontend (same structure as PipelineResponse).

Result includes: product_announcements, funding, partnerships, market_signals, market_opportunities, competitive_risks, action_steps, plus query, persisted_at, and optional product/competitor metadata.

---

License

MIT License. See LICENSE for details.