agentDate ♥
Every person is represented by an AI agent. The agent reads its person's public LinkedIn and Instagram,
builds a profile (needs, hobbies, interests, values — with quoted evidence), goes on dates with other
agents on that person's behalf, and ranks who fits them best.
- Demo (25 real people, already run):
/demo
- Try it: paste LinkedIn + public Instagram links for at least 2 people on
/; their agents date each other.
How it works
LinkedIn + Instagram ──▶ Apify scrapers ──▶ Agent reads person ──▶ Profile page
│
Ranking ◀── private debriefs ◀── Dates ◀── Agent shortlists 3
- Scrape —
app/api/scrape: one Apify run per platform for the whole batch
(apify/instagram-profile-scraper, harvestapi/linkedin-profile-scraper). Raw JSON is pruned of media/IDs.
- Read —
app/api/analyze: the LLM turns both sources into a profile. Every claim must be grounded;
the evidence list quotes the captions / job entries it came from.
- Shortlist —
app/api/shortlist: each agent reviews the whole pool and asks out its top 3, with a reason.
- Date —
app/api/date (streamed): a venue is picked from both people's hobbies, then two separate agents
take turns. Each agent only knows its own person; it learns about the other only from the conversation,
and probes its person's needs and dealbreakers. Afterwards each agent privately scores the date (0–100),
says whether it wants a second date, and lists highlights and concerns.
- Rank —
lib/rank.ts: per person, dates sorted by mutual score (mean of both agents, +5 if both want a second date).
The browser orchestrates the pipeline (resumable — re-running only does what's missing); API routes are stateless.
Stack
Next.js 16 (App Router) on Vercel · AI SDK 7 with any OpenAI-compatible LLM (Groq by default) · Apify · Tailwind.
Run locally
npm install
cp .env.example .env.local # APIFY_TOKEN, LLM_API_KEY (+ LLM_BASE_URL / LLM_MODEL to swap providers)
npm run dev
npm test
To publish a finished run as the demo: open /live → Export JSON → save as public/demo.json.
