Solisa AI — About the Project Why we built it (Inspiration)
Insurance sales has tons of intent but even more friction. Leads fill a form, and then… nothing: context gets lost across emails, texts, and calls; follow-ups are late or generic; and booking time on a calendar is another hurdle. We wanted an AI SDR that acts the way a great rep does—fast, contextual, consistent—so a visitor can go from form fill → personalized outreach → booked meeting in minutes, with no manual lift.
What it does (TL;DR)
Solisa AI captures a lead, enriches them, crafts hyper-personalized outreach (SMS, email, LinkedIn), and books meetings—then keeps following up agentically using call transcripts, texts, and emails to decide the next best action.
Phase 1 – AI Lead Gen: Capture → enrich via Clay webhook → generate personalized SMS/Email/LinkedIn with GPT-4 → send email (console or SMTP) with Calendly link → show everything in a clean Next.js UI.
Phase 2 – Agentic Follow-ups: Ingest long-form context (calls, threads, CRM notes), track objections/intent shifts, and automatically propose/send the next step (text, email, call script, or wait). Escalates to a human with a one-page brief.
How we built it (Stack & Architecture)
Frontend
Next.js 16 (App Router) + Tailwind for a modern, responsive UI.
Pages: / (landing), Leads Dashboard, Capture form, Personalize (per-lead AI messages), Autopilot (next best action).
Components for theme toggle, email preview modal, and confirmation flows.
Environment-driven: NEXT_PUBLIC_API_BASE and NEXT_PUBLIC_CALENDLY_URL.
Backend
FastAPI (Python) with SQLAlchemy (PostgreSQL) for a reliable, typed API.
Clay enrichment (async webhooks): leads are created fast, enriched later via a secure callback (x-callback-token).
OpenAI GPT-4(o-mini) via openai + httpx for low-latency personalization.
Messaging
Email: console/EML outbox for demo (or SMTP drop-in).
SMS: Twilio “dry-run” provider + inbound webhook simulators.
Agentic brain endpoints:
POST /api/leads/{id}/followups/ingest — store transcripts/threads.
POST /api/leads/{id}/followups/autopilot — compute next best action from the combined timeline.
Data model (simplified)
Lead: core attributes + enrichment fields (company, title, size, industry, LinkedIn).
Message: outbound/inbound items (SMS/email), status, timestamps.
FollowupContext: normalized artifacts (transcript snippets, notes, objections) keyed to a lead.
Built With
- docker
- gpt
- javascript
- llm
- openai
- postgresql
- python
- react
- sql
Log in or sign up for Devpost to join the conversation.