FlyQualify AI — Autonomous Inbound Revenue Intelligence Platform

Hackathon submission for FlytBase — Inbound BDR / AI-Native Qualifier track.

FlyQualify AI ingests a raw inbound lead (email / contact form text) and, through a visible six-agent AI pipeline, produces an AE-ready opportunity record: parsed facts, an intelligence brief on the company, an explainable BANT + MEDDPICC score, the closest FlytBase reference deployment, a personalised 3-email outreach sequence + LinkedIn + meeting invite, and a Salesforce-grade AE handoff card — all in ~20 seconds.

It is designed to look and feel like real internal enterprise software (Linear / Attio / Salesforce Einstein), not a chatbot.

---

Architecture

flowchart LR
  U[Inbound email / form text] --> A1
  subgraph Pipeline[FastAPI /api/qualify-lead]
    A1[Agent 1<br/>Lead Parser] --> A2[Agent 2<br/>Company Research]
    A2 --> A3[Agent 3<br/>Qualification<br/>BANT + MEDDPICC]
    A3 --> A4[Agent 4<br/>Case Study Match<br/>+ GTM Motion]
    A4 --> A5[Agent 5<br/>Outreach Generator]
    A5 --> A6[Agent 6<br/>AE Handoff]
  end
  A6 --> Mongo[(MongoDB — fq_leads)]
  A6 --> UI[React dashboard<br/>Sidebar-navigated results]

Loading

Every agent is its own Python module in backend/agents/ with:

• its own system prompt,
• a strict JSON output schema enforced in the prompt,
• a deterministic fallback so the app never breaks if the LLM key is absent or rate-limited.

Runs are persisted to MongoDB so the History page reads a real record, not local state.

---

Tech stack

• Frontend — React 19 + React Router, Tailwind, shadcn/ui, Framer Motion, Recharts, lucide-react, sonner.
• Backend — FastAPI (Python 3.11), Motor (async MongoDB driver).
• AI — OpenAI GPT-4o-mini via the Emergent Universal LLM Key (emergentintegrations). Model + provider are single env vars so they can be swapped instantly.
• Web research — LLM-estimated intelligence (by design — no external search dependency required for the demo). The agent output is honest about being an estimate.
• Storage — MongoDB collection fq_leads with a UUID string _id.

---

Environment variables

Var	Purpose	Fallback if absent
MONGO_URL	MongoDB connection string	required
DB_NAME	Mongo database	required
CORS_ORIGINS	Comma-separated allowlist	*
EMERGENT_LLM_KEY	Universal LLM key	App still works — every agent falls back to a deterministic estimate
LLM_PROVIDER	openai | anthropic | gemini	openai
LLM_MODEL	Model name	gpt-4o-mini
REACT_APP_BACKEND_URL	Frontend → backend base URL	required

Zero-key demo path: with EMERGENT_LLM_KEY unset, every one of the six agents falls back to a fully-populated deterministic result. The full UI works end-to-end with no keys configured.

---

Local development

# Backend
cd backend
pip install -r requirements.txt
uvicorn server:app --reload --host 0.0.0.0 --port 8001

# Frontend
cd frontend
yarn install
yarn start

Then open http://localhost:3000, load a sample lead, and click Run AI qualification.

---

API surface

Method	Path	Description
GET	/api/sample-leads	3 built-in inbound lead samples
GET	/api/case-studies	Case study library used by Agent 4
POST	/api/qualify-lead	Runs the 6-agent pipeline. Body: { "raw_text": "…" }
GET	/api/leads	List of historical runs (summary)
GET	/api/leads/{id}	Full run detail

---

Deployment

The app is provisioned on the Emergent platform (FastAPI + MongoDB + React, /api prefix ingress) and is one-click deployable via the Emergent deploy button. The same code can also be shipped to Vercel (frontend) + Render/Railway (backend) since it is a standard FastAPI + React project.

---

Screenshots

• Home / input
• Live pipeline (six agents animating sequentially)
• Research view
• Qualification (BANT + MEDDPICC)
• Case Study & GTM motion
• Outreach (email tabs + LinkedIn + meeting invite)
• AE Handoff (Salesforce-style)
• History