User repository on GitHub
Real-time hawk-or-dove analysis of Fed press conferences with TypeSafe's Jev vs chat models. Live at fedspeech.pages.dev
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
Jev directory
CommunityJev Demos & Experiments
Jev vs Chat Model: Hawk or Dove?
Real-time hawk-or-dove analysis of Fed press conferences with Jev side by side with chat models. Jev scores every word the Chair says in about 150 ms while the chat model scores each caption line in a few seconds, both judged against the same four stances.
- Stars
- 1
- Language
- JavaScript
- Reading time
- 3 min
- Published by
- Community
- Author
- turboguo
- Last verified
- Oct 2026
Used inJev for Finance & Fintech
- Languages
- JavaScriptTypeScript
- Platforms
- Cloudflare
- Capabilities
- Open source
Highlights
- Jev scores every word the Chair says in about 150 ms; the chat model scores each caption line in a few seconds, judged on the same four stances.
- Three modes: a recorded July 29, 2026 press conference demo with real answers and timings, past conferences by date, and type-or-paste judging.
- One Cloudflare Worker: public/ is the page and src/ is the API; a Pages site forwards requests over a service binding so visitor IPs reach the rate limits.
- Themes are typed questions (noul, choice, score) in src/standards.ts; chat models get the same standard as a JSON prompt.
- Past conferences save answers, so each is paid for once and replays free; Fed captions appear a few days after the event.
Quickstart
bash
npm run dev
npm run record-demo -- --fed-date 2026-07-29 --model opus-5Watch out
Copy reviewed against the repo README on 2026-10-15; no accuracy benchmarks are published for either side, and the digest truncates the README's payment section mid-sentence. Fed data comes from the same public requests the Brightcove player makes, which the README notes is undocumented and could change.
Questions about this entry
Short answers drawn from the linked project or page.
A live site that judges whether the Fed Chair sounds hawkish or dovish during press conferences, running TypeSafe's Jev side by side with a chat model. Jev scores every word in about 150 ms; the chat model scores each caption line in a few seconds.
Watch the recorded demo: a replay of the July 29, 2026 press conference with Chairman Warsh using real answers and real timings, making no API calls. Past conferences also save their answers, so each is paid for once and replays free.
Claude Opus 5, GPT-5.6 Sol, GLM-5.3 and DeepSeek V4 Flash. They run on the site's account through Cloudflare AI Gateway with Unified Billing, or with a visitor's own key from Anthropic, OpenAI, Z.ai or DeepSeek. The list lives in src/models.ts.
Hawkish, Partly hawkish, Partly dovish and Dovish. Both Jev and the chat models are judged against the same four, and the same standard is given to the chat model as a prompt that must be answered in JSON, per STANDARDS.md.
The Worker makes the same public requests the Fed's Brightcove player makes and fetches the video's English caption file — the official transcript, timed and labeled by speaker. The README warns this is not a documented API, so it could change, and captions usually appear a few days after the event.
By default the site does: Jev on the site's TypeSafe setup, and the chat models through a Cloudflare AI Gateway with Unified Billing, so the site holds no provider keys. You can also bring your own provider key for Anthropic, OpenAI, Z.ai or DeepSeek.
With npm run dev running, run npm run record-demo -- --fed-date 2026-07-29 --model opus-5; the demo is stored as public/demos/fed-demo.json. The deployment is a Cloudflare Worker with a Pages site in front of it.
More Jev resources like this
Watches the Bluesky firehose and asks Jev eight questions about every sampled post in one call, covering intent, topic, sentiment, engagement bait, hostility, sarcasm, bot-ness and screen safety, then routes anything the model will not commit to into a needs-a-human lane. Live at cloutmetrics.ai.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
ragelink/firehose-judgeTypeScript
ragelink/firehose-judgeTypeScriptStreamlit demo that turns live market data into structured investment decisions with TypeSafe Jev: yfinance supplies prices and fundamentals, then a 20-question model scores buy/sell conviction, financial health and risk.
Jev Demos & Experiments#community#catalog-import#logicrw
Community
0xzee/jev-stock-decision-makerPython
0xzee/jev-stock-decision-makerPythonA browser chess coach where Stockfish (WASM) owns the truth about the position and Jev makes the coaching judgment after every move, with templates and a small writing model supplying the words. Status: Wave 0 scaffold and Jev feasibility spike.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
joellewis/game-coachTypeScript
joellewis/game-coachTypeScriptA live, graphical dojo for Jev with six real-time canvas scenes over one WebSocket: task routing across model tiers, ticket triage, inbox-scale queues and an honest Jev-vs-Claude gauntlet. Boots in SIM mode without an API key.
Jev Demos & Experiments#community#catalog-import#madewithjev
A local app that turns a worded query into a physics pile of floating YC logos: retrieval over 6,241 companies by meaning, logo image, OCR'd logo text, and Jev re-applied tags, with Jev scoring the finalists into calibrated probabilities.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
aayan-dev/aayans-yc-indexorHTML
aayan-dev/aayans-yc-indexorHTMLA small local app that turns Jev's per-pixel probability distributions into paintings: stroke pigments are sampled from the probabilities, mean color supplies the underpainting, and entropy controls relief, rendered in a browser worker.
Jev Demos & Experiments#community#catalog-import#madewithjev
Community
achimala/jevinciJavaScript
achimala/jevinciJavaScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.