🗺️ Development Roadmap

← Home

---

✅ Completed

Phase 1 — Foundation

• Playwright persistent Chrome context (reuses real session)
• LinkedIn company search + profile scraping
• AI scoring (keyword + Ollama/Gemini)
• Connection request automation with personalized note
• Daily/weekly cap enforcement
• FastAPI backend with WebSocket log streaming
• Real-time dashboard UI

Phase 2 — First Message

• Msg Prospect tab with messages-per-run slider
• Connection card scraping via JS DOM evaluation
• [data-view-name="message-button"] selector (confirmed from Web Inspector)
• Direct compose URL navigation (more reliable than clicking button)
• Ollama message generation (warm, non-salesy prompt)
• Deduplication against messages.csv
• 5-second send countdown with abort
• LinkedIn Premium AI prompt dismissal

Phase 3 — Follow-Up Sequence

• followups.csv schema + CRUD (upsert_followup, seed_followups_from_messages)
• Follow Up tab with days slider + cap slider
• Reply detection via message bubble CSS classes
• Profile page Message <button> click (not <a> tag)
• Compose overlay opens on profile page (no page jump)
• AI follow-up generation referencing original message
• Status tracking: pending → followed_up / replied
• AI quote stripping (.strip('"').strip("'"))

Phase 4 — Hardening

• Git history cleanup (PII purged via filter-branch)
• .gitignore pattern-based (*.csv, *.log, *.jsonl)
• messages.csv recovery from git stash
• README updated (no Sales Navigator references)
• messenger_debug.log file-based verbose logging

---

🔲 Planned — Near Term

Smart Follow-Up Timing

• Respect wait_days per-person (already built, just set slider > 0)
• Skip weekends when calculating wait (B2B best practice)
• Configurable follow-up window (e.g. only send Mon–Thu 9am–5pm)

Reply Handling

• When replied detected: log reply timestamp to followups.csv
• UI: show "Replied ✓" rows in green in the Follow-Up table
• Optional: notify user (desktop notification or email) when someone replies

Message Quality

• A/B testing: rotate between 2 message templates, track which gets more replies
• Message preview before send (show in UI, require click to confirm)
• Per-company custom message context (e.g. "mention their recent Series B")
• Tone selector: Professional / Casual / Direct

---

🔲 Planned — Medium Term

Analytics

• Reply rate tracking (replied / followed_up)
• Connection acceptance rate
• Best-performing message templates
• Time-of-day send analysis
• Simple dashboard charts (Chart.js)

Campaign Management

• Named campaigns (group companies + message templates together)
• Campaign start/stop/archive
• Per-campaign stats (sent, replied, converted)

Multi-Step Sequences

• Step 2 follow-up (after follow-up also gets no reply)
• Sequence designer: define N steps with N-day gaps
• Hard stop: never contact someone more than 3 times total

---

🔲 Planned — Longer Term

Profile Intelligence

• Pull profile headline, recent posts, mutual connections before messaging
• Feed profile context into AI prompt for hyper-personalization
• Detect job changes: "Congratulations on the new role at..."

Safety & Rate Limiting

• Randomized send times (not all at :00 or :30)
• Auto-pause if LinkedIn shows any warning banners
• Monthly cap tracking (LinkedIn restricts ~100 connection requests/week)
• Exponential backoff on repeated CAPTCHA

Export & Integrations

• Export followups.csv to Notion database
• Export to Google Sheets (via gspread)
• Webhook on reply detected (Zapier / Make compatible)
• Slack notification on reply

---

💡 Ideas Backlog