Get Qualified Leads From Luma

Get Qualified Leads from Luma

Chains luma-event-attendees (search mode) → timeframe filter/dedup → lead-qualification → sheet export → Slack alert. Both Apify and Slack are optional: with Apify you get full guest profiles (recommended), else a keyless Luma event-page scrape (hosts + metadata); with Slack creds the alert is delivered, else it is written to a file.

When to use

• "Find qualified leads from events." / "Who's attending [topic] events that match our ICP?"
• "Prospect event attendees and alert me with the top ones."

How to run

Step 1 — clarify params (you, the agent)

Confirm location, topics (3-5 variations), timeframe, ICP/qualification prompt, Slack target, and top N. Always confirm a timeframe — Luma returns events from all time.

Step 2 — parallel Luma searches

# run per topic+location variation (Apify Luma actor; requires APIFY_API_TOKEN)
python3 ${SKILL_DIR}/scripts/luma_search.py --search "AI agents San Francisco" \
  --output ${WORKSPACE}/s1.json
python3 ${SKILL_DIR}/scripts/luma_search.py --search "LLM apps SF" \
  --output ${WORKSPACE}/s2.json
# ... 3-5 variations

Keyless degrade (no APIFY_API_TOKEN): luma_search.py requires Apify (full guest profiles). Without it, collect candidate event URLs (agent web search) and direct-scrape each with the keyless Playwright scraper — hosts + metadata only, not the full guest list:

npx playwright install chromium   # one-time
node ${SKILL_DIR}/scripts/luma_scrape_pw.mjs \
  --event-url https://lu.ma/abc123 --output ${WORKSPACE}/s1.json

Step 3 — filter timeframe + dedup

python3 ${SKILL_DIR}/scripts/filter_dedup.py \
  --inputs ${WORKSPACE}/s1.json ${WORKSPACE}/s2.json ${WORKSPACE}/s3.json \
  --since-days 30 --output ${WORKSPACE}/people.json

Filters events by event_date, merges, dedups by lowercased name (skips null names).

Step 4 — qualify (sub-skill)

Run lead-qualification over people.json with the ICP prompt → verdict + score. Write the qualified set to qualified.json.

Step 5 — export + Slack alert

Export qualified.json to a sheet / channel attachment, then:

python3 ${SKILL_DIR}/scripts/slack_alert.py \
  --leads ${WORKSPACE}/qualified.json --top-n 5 --channel "#leads"

Uses SLACK_WEBHOOK_URL (webhook) or SLACK_BOT_TOKEN + --channel. Keyless degrade (no Slack creds): the alert is written to --out-file (default luma_alert.txt) and printed to stdout, so the pipeline still produces a deliverable for the agent to relay.

Outputs

• people.json — merged, timeframe-filtered, deduped attendees.
• A qualified lead sheet (attendee, event, company, verdict/score) + channel attachment.
• A Slack alert listing the top N leads.

Credentials / env

• Required: none. The pipeline runs keyless via the Playwright direct-scrape + file-based alert.
• Optional:
  • APIFY_API_TOKEN — if set → Apify Luma actor search mode (full registered-guest profiles: LinkedIn/X/bio — recommended). If not → keyless luma_scrape_pw.mjs direct-scrape (hosts + metadata only).
  • SLACK_WEBHOOK_URL or SLACK_BOT_TOKEN — if set → Slack delivery. If not → the alert is written to a workspace file (--out-file) and printed.
  • APIFY_LUMA_ACTOR (actor override); ANTHROPIC_API_KEY only if the qualification LLM isn't platform-provided; Google-sheet creds if not platform-provided.

Notes & edge cases

• Always confirm a timeframe — filter by event_date to avoid stale leads.
• Run keyword variations in parallel for coverage, then dedup once merged.
• Handle null/None attendee names gracefully (skipped).
• Direct Luma scrape (no Apify) yields hosts + metadata only, not full guest profiles — prefer the Apify search mode when full registered-guest coverage matters.
• The Playwright degrade needs npx playwright install chromium once.