DeloConnect — AI Employee Wellness Platform
Deloitte OpenSoft 2025
An AI system that reads the early warning signs of employee burnout — before HR has to ask.
Overview
DeloConnect was built for Deloitte OpenSoft 2025, where it won a Gold Medal. It's an AI employee-wellness platform centered on a LangGraph fan-out/fan-in workflow: five specialized agents analyze activity, leave, performance, and sentiment signals in parallel, then converge into structured reports for HR — alongside a counseling chatbot employees can talk to directly.
The Problem
HR teams usually learn about burnout or disengagement far too late — once someone has already filed a complaint or resigned. The early signals exist (declining activity, irregular leave patterns, dropping performance, low sentiment) but they're scattered across systems and never connected.
Approach
Fan-out/fan-in multi-agent workflow
Built a LangGraph workflow where 5 specialized agents — covering activity, leave, performance, rewards, and more — run in parallel (fan-out) over employee data, then converge (fan-in) into a single structured HR report per employee.
Counseling chatbot with a decision agent
Designed a counseling chatbot backed by an Agno/OpenAI decision agent that monitors the conversation in real time, deciding when to shift topics, escalate to HR, or end the session — so employees get support without every conversation needing a human.
Automated escalation reports via Gemini
When a chat is escalated, a Gemini-based summarizer converts the full conversation into a structured HR report — flagging key concerns, sentiment trends, and recommended action items, so HR can act on a digest instead of a transcript.
Early at-risk detection model
Engineered an Isolation Forest ensemble over EMA-smoothed activity, leave, and Vibemeter signals to flag at-risk employees statistically, ahead of any explicit complaint.
Secure, real-time backend
Built the backend on FastAPI and MongoDB Atlas with JWT authentication and WebSocket-based chat, enforcing strict role-based access control between HR and employee views.
Results
- Won a Gold Medal at Deloitte OpenSoft 2025
- 5-agent fan-out/fan-in pipeline turns raw HR signals into structured, actionable reports automatically
- Isolation Forest ensemble flags at-risk employees from EMA-smoothed multi-signal data, ahead of escalation
- Real-time WebSocket counseling chat with automatic HR escalation and strict RBAC
Stack
More projects