InsourceCrew

• 

  Support Employee output

• 

  Support Employee output

• 

  Sales Employee output

• 

  Operations Employee output

• 

  Recruiter Employee output

• 

  Marketing Employee output

• 

  Usage Guide Page

• 

  Marketing Employee output

• 

  Marketing Employee output

• 

  Marketing Employee

• 

  Operations Employee

• 

  Landing Page

• 

  My Crew Page

• 

  Usage Guide Page

• 

  Login Page

Inspiration

Small businesses have too much work and not enough people. We wanted to make AI feel like a helpful teammate, not another complicated tool. InsourceCrew lets businesses hire AI employees to help with everyday work like hiring, support, sales, marketing, and operations.

What it does

InsourceCrew lets a business hire five AI employees: Recruiter, Support, Sales, Marketing, and Operations. Each employee has its own versioned workflow that users can inspect, edit, publish, restore, and run. Users upload realistic business files such as job descriptions and resumes, support tickets, lead records, campaign briefs, or operational reports. The AI employee processes the data with GPT-OSS and returns a structured, role-specific business deliverable directly in the app. For example, the Recruiter evaluates candidates and recommends next steps; Sales qualifies leads using BANT and creates outreach; Support classifies tickets and drafts responses; Marketing creates campaign plans; and Operations identifies bottlenecks and improvement actions.

How we built it

Frontend: Next.js, TypeScript, Tailwind CSS, React Flow, Framer Motion Backend: FastAPI, SQLAlchemy, Pydantic, JWT authentication Database: PostgreSQL on Supabase AI: GPT-OSS-120B through Fireworks AI Deployment: Vercel for the frontend and Render for the backend

Challenges we ran into

The largest challenge was making the product feel like real AI work rather than a static demo. We needed uploaded files to become useful context for each employee, while ensuring the output was specific to the employee’s role. We also solved challenges around workflow version isolation, structured LLM output, frontend rendering of variable JSON responses, deployment configuration, CORS, and connecting Vercel, Render, Supabase, and Fireworks reliably.

Accomplishments that we're proud of

Built five distinct AI employees with role-specific inputs, workflows, and outputs. Created editable, versioned workflows with draft, publish, restore, and history capabilities. Made GPT-OSS process actual uploaded company files instead of relying on static mock text. Built session history so every employee run is recorded with steps, timing, and deliverables. Created polished, structured business reports instead of exposing raw JSON to users. Delivered a deployed full-stack product with authentication, database persistence, and an AI execution engine.

What we learned

We learned that strong AI products need more than an LLM call. They need reliable input handling, structured outputs, clear workflow boundaries, useful UI feedback, and an experience that helps non-technical users understand what the AI did.

What's next for InsuranceCrew

We learned that strong AI products need more than an LLM call. They need reliable input handling, structured outputs, clear workflow boundaries, useful UI feedback, and an experience that helps non-technical users understand what the AI did.

Next, we plan to add real integrations with tools such as Gmail, Slack, calendars, CRMs, help desks, and cloud storage. We also want to introduce organization-specific knowledge bases, scheduled and event-triggered employee runs, approvals for high-risk actions, team collaboration, and deeper analytics. The long-term vision is for InsourceCrew to become a practical AI workforce platform where businesses can hire, customize, supervise, and scale AI employees across their daily operations.

Built With

• 5.6
• codex
• fastapi
• nextjs
• postgresql
• suppabase
• terra
• vercel