🚀 HRishta – An AI Assistant for Smarter Human Resource Management

💡 Inspiration

The idea for HRishta was born from observing how overloaded HR departments often are with repetitive tasks—responding to FAQs, managing leave requests, shortlisting resumes, and conducting initial interviews. Despite the shift to digital HR platforms, a lot of HR work still requires manual effort. We wanted to build an AI assistant that acts like a trusted HR team member, helping streamline operations while preserving the human touch in "Human Resources".

The name HRishta is a play on "HR" and the Hindi word "Rishta" (relationship), symbolizing the connection between employees and organizations.

🧠 What I Learned

• Agentic AI Design using Google ADK

• Designing modular sub-agents for specific tasks like:

  • FAQ answering (RAG-based)
  • Resume shortlisting (BERT/spaCy)
  • Interview QnA + Evaluation (LLM + Keyword Matching)
  • Leave management with holiday + weekend logic
  • Email automation via Gmail API

• Integrating FastAPI + Firestore (GCP) for a scalable, real-time backend

• Understanding leave policies and how holidays/Sundays are counted in real organizations

• How to route user inputs through a central agent using intent classification

🛠️ How I Built It

🔨 Tech Stack

• Frontend: React (planned for HR dashboard)
• Backend: FastAPI on Google Cloud Run
• Database: Firestore (NoSQL) for employee, leave, and resume data

• AI Components:

  • FAQ Agent: mistral + haystack (RAG-based)
  • Resume Parser: spaCy + rule-based classifiers
  • Interview Evaluator: LLM + scoring rubric

• Agent Framework: Google ADK for structured agent development

🧩 Agent Design

MainHRAgent
├── FAQAgent          (RAG + vector DB)
├── EmailAgent        (SMTP/Gmail API)
├── ResumeShortlister (NLP + scoring)
├── InterviewAgent    (LLM QnA + eval)
└── LeaveAgent        (Calendar-aware)

Each agent implements a run(input: str) -> str method and can be queried via an intent-routing system.

⚔️ Challenges I Faced

• Leave calculation logic: Handling weekends and national holidays while applying leaves accurately.
• Resume parsing & shortlisting: Creating fair and bias-free rules while keeping results meaningful.
• Conversation routing: Ensuring that the main agent correctly identifies user intent and calls the right sub-agent.
• LLM evaluation: Automating first-round interviews required building a reliable keyword matching + grading strategy.
• Deployment and Security: Making sure sensitive employee data is securely stored and only visible to HR.

✅ Final Thoughts

HRishta is more than just a chatbot. It’s a productivity companion for HR teams — built to reduce repetitive tasks and let humans focus on what they do best: building relationships. We're excited to keep improving it with features like document handling, voice QnA, and integration with existing HRMS tools.

"An HR department powered by AI shouldn't feel robotic — it should feel more human. That’s what HRishta aims to achieve."

Built With

• adk
• python