Agentic AI Onboarding

Inspiration

The project aims to rethink and streamline the new-hire onboarding process by combining HR workflows, interactive portals, and AI assistance. As the README puts it: “An intelligent onboarding system that combines HR management, employee portal, and AI-powered assistance with automated email routing.”
The motivation appears to be to reduce friction for:

HR administrators managing many hires
New employees grappling with setup, tasks, integrations
The delay or manual burden of provisioning access, answering basic questions, routing requests

By bringing in an AI assistant (avatar/chatbot) and automated routing, the project is inspired by the idea that onboarding should be proactive, conversational, and low-friction.

⸻⸻⸻⸻⸻⸻

What it does

Here are the key functional features:

-HR Admin side: quick employee registration, dashboard of all onboarded employees and status, automatic provisioning of credentials.

-Employee side: Personalized portal login, interactive onboarding checklist, view of connected services (Slack, GitHub, etc).

-AI Assistant: Text or voice avatar chat; the assistant can understand requests like “I need access to Adobe Creative Cloud”, identify the department/need, and automatically send emails to the appropriate group (IT, HR, manager) via an email API.

-Tech stack: Frontend built with Next.js 16 + React 19 + TypeScript; Tailwind CSS for styling; Supabase (PostgreSQL) backend; Authentication via Supabase; Email via Resend API; AI via “Anam AI” with GPT-4.1 & function calling.

-The architecture includes API Routes: employees/create, auth/login, send-email, chat.

-Database schema (example) for employees: id, email, password, name, role, project, manager_email, onboarding_status, created_at.

-The AI assistant uses defined functions (send_email_to_it, send_email_to_hr, send_email_to_manager) to handle routing when appropriate.

⸻⸻⸻⸻⸻⸻

How we built it

Key building steps and architecture:

-Selected Next.js (React + TypeScript) for the frontend, enabling component-based UI and serverless API routes.

-Used Supabase for database + authentication (quick setup for demo).

-Used Resend API for sending transactional emails.

-Integrated Anam AI (GPT-4.1) with function calling support: the assistant listens to user input, infers intent, and triggers appropriate actions (email routing) via API.

-The UI components include: login form, HR onboarding dashboard, employee dashboard, dual-chatbot interface (text & avatar), onboarding checklist, integrations panel.

-Environment variables (.env.local) configure Supabase credentials, Resend API key, Anam API key, avatar/voice IDs for the AI.

-Demo/test flow: HR creates employee -> employee logs in -> uses AI avatar chat -> system sends email via Resend based on request.

-The project is marked as a “demo application with simplified security” and notes that in production you’d need proper password hashing, JWT auth, input sanitisation, rate limiting, secure headers, etc.

⸻⸻⸻⸻⸻⸻

Challenges we ran into

-Some inferred challenges (since the README doesn’t explicitly list many “we hit this” items):

-Security and production-readiness: The README flags that it’s a demo and alerts that proper security measures are needed (password hashing, JWT, input sanitisation). This suggests that moving from demo to production is non-trivial.

-Intent recognition and function calling: Mapping natural language from a user (voice or text) to the correct internal function (send_email_to_it / HR / manager) reliably is a challenge in any AI-assisted workflow. Ensuring the AI extracts correct parameters (employee name, request type, details) and routes properly is likely a non-trivial part.

-Multi-channel interface: supporting both text-based and avatar/voice chat (microphone, avatar ready state) adds UI/UX complexity (e.g., handling permission for mic, avatar loading). The troubleshooting section mentions “Avatar not loading: check microphone permissions, verify Anam API keys, check browser console errors”.

-Onboarding checklist and integrations: managing the state of “onboarding_status”, handling tasks, tracking progress, showing connected services implies complexity in front-end state, backend sync, user flows.

-Automated email routing: Ensuring that email templates, correct recipients, correct context, at correct time (and not sending duplicates or sending inappropriate emails) is a domain challenge.

-Building such a product end-to-end (UI, backend, AI, routing) within a demo scope means managing scope, ensuring the prototype is coherent, but not overshooting into full enterprise-grade.

⸻⸻⸻⸻⸻⸻

Accomplishments that we’re proud of

-Integration of AI (Anam AI + GPT-4.1) with “function calling” to automate meaningful real-world onboarding tasks (e.g., routing an IT access request). That’s a strong proof-of-concept.

-A full stack solution: from HR admin UI to employee portal to AI chat interface to backend email sending — shows a comprehensive end-to-end flow rather than a narrow demo.

-Use of modern stack (Next.js, TypeScript, Supabase, Tailwind) meaning the prototype is current and maintainable.

-Clear architecture and documented setup: README includes Quick Start, Setup Instructions, Implementation Summary, Demo Guide etc. The fact that one can set up environment variables, run dev server, and test the AI avatar flow shows good demo maturity.

-The onboarding checklist & integrations panel show awareness of the real elements new employees face: not just login, but tasks, service integrations, AI assistance.

-The demo scenario emphasises employee-centric UX: “I need Adobe Creative Cloud”, “I have a question about vacation policy”, “How do I contact my manager?” – the choice of example requests shows user empathy.

⸻⸻⸻⸻⸻⸻

What we learned

-That onboarding workflows can be significantly improved via conversational AI + automation: rather than employees figuring out who to email or where to ask, the system can intercept natural language and route appropriately.

-That the “last mile” of employee onboarding (task checklists, integrations, quick answers) is often underserved, and designing for that can reduce friction.

-That building a demo is feasible with modern stacks (Next.js, Supabase, APIs) and AI tools – enabling rapid iteration.

-But also that productionising this sort of system requires serious attention to security (passwords, authentication, sanitisation), scalability (many employees/requests), and reliability of AI intent mapping.

-That combining UI/UX, backend, AI, and business process workflow demands an interdisciplinary approach (designers, engineers, AI experts, onboarding subject matter experts).

-That articulating clear API surfaces (employees/create, chat endpoint, send-email endpoint) and mapping function calling is critical for developer clarity and maintainability.

⸻⸻⸻⸻⸻⸻

What’s next for Agentic AI Onboarding

-Production-Readiness: Implement proper security (password hashing, JWTs, rate limiting), scalability, and perhaps multi-tenant support for organisations.

-Expand the AI capabilities: beyond simply routing emails, perhaps to proactive suggestions (e.g., “You haven’t completed your license request, want me to take care of it?”), training/learning path generation, onboarding analytics.

-Deeper integrations: connect with more services (Slack, GitHub, Azure AD, identity access management) so the system can not only route emails but trigger provisioning workflows, monitor completion of integrations.

-Personalisation: tailor the onboarding flow for role, team, department — more intelligent checklists, adaptive help based on the employee’s profile.

-Metrics & analytics: track key onboarding KPIs like time-to-first-value, completion rate of onboarding tasks, employee satisfaction (NPS/eNPS). This aligns with HR metrics best practices.

-Multi-modal interaction: improve voice/avatar UX, support for mobile, support for accessibility.

-Internationalisation/localisation: support for multiple languages, cultures, compliance requirements.

-Learning & development extension: the onboarding system could evolve into a continuous learning/engagement platform beyond the first days/weeks of employment.

-Open-sourcing/handoff: making the codebase easier for organisations to adopt, extend, and customise for their own domains.