Ticker research with parallel web search and LLM synthesis. Built for the CascadiaJS 2026 workshop on Render Workflows.
Enter a ticker or any research question. The app runs four Exa web searches in parallel, then asks Claude to write a structured research memo from the results. Version 1 is intentionally flaky: each search has a 30% random failure rate so the workshop can show what breaks before adding Render Workflows.
This app is meant to run on Render. Local setup is optional (see Developing below).
- Click Deploy to Render (or apply the
render.yamlBlueprint from the repo). - In the Dashboard, set environment variables:
| Variable | Required in v1 |
|---|---|
ANTHROPIC_API_KEY |
Yes |
EXA_API_KEY |
Yes |
RENDER_API_KEY |
Reserved for v2 |
- Wait for the deploy to finish. Open your service URL (for example
https://ticker-research.onrender.com). - Enter a query and run a research pass. Most v1 runs fail on purpose; that is expected until v2.
Health check: GET /healthz returns ok.
TODO: add static/images/architecture-diagram.png and static/images/pipeline-flow.png.
tasks/src/search.ts throws on roughly 30% of calls with a fake rate-limit error. Four parallel searches via Promise.all in tasks/src/research.ts means most v1 runs fail when any single search throws. That is the workshop teaching moment. Version 2 wraps the same task files in Render Workflows with retries.
| Path | Purpose |
|---|---|
server/src/ |
Express API and SSE stream |
tasks/src/ |
Search, synthesis, and orchestration (becomes Workflow tasks in v2) |
ui/src/ |
React UI |
shared/types.ts |
Shared event and data types |
Use this only if you are changing code before deploy: workshop prep, or contributing.
git clone https://github.com/ojusave/workshop-demo.git
cd workshop-demo
npm install
cp .env.example .env # ANTHROPIC_API_KEY and EXA_API_KEY
npm run devUI: http://localhost:5173 (proxies /api to the server on port 3000).
Generated memos are for demonstration only. They are not financial advice.