ConvoHire - Agentic Recruiter

A modular, LLM-powered interview system that reads a job description and a candidate resume, then conducts a five-question technical interview with smart follow-ups, real-time feedback, and backend evaluation.

---

✨ Features

• 🔍 Parses resume & job description
• 🧠 Enhanced skill extraction from JD/Resume and matching using semantic similarity cosine threshold for finding exact and very similar matches, powered by Gemini 2.5 Flash.
• ❓ Technical interview questions powered GPT-4o-mini(or any llm of your choice) varying with the depth of candidate's experience.
• 🔁 Contextual follow-up questions based on candidate’s answers
• 📝 Real-time feedback with hidden scoring (stored server-side)
• 🗂️ Session memory using FastAPI (easily extendable to Redis)
• 💬 Interactive frontend with Streamlit chat interface
• ⚡ Low LLM usage per interview (1 generate + ≤5 eval + optional follow-ups)

---

Technical Flow

Agentic-Recruiter (ConvoHire) – Technical Flow Brief

Overview: ConvoHire is a modular, LLM-powered interview system that simulates human-like technical interviews. It compares a candidate's resume with a job description (JD), extracts and matches skills, and conducts a dynamic question-answer session with backend evaluation and real-time feedback.

Main Technical Flow:

• Input and Parsing:

  • The system takes a candidate resume and a job description as input.
  • Uses custom parsers (resume_parser.py, jd_parser.py) to extract relevant skills and details from both documents.

• Skill Extraction & Matching:

  • Extracted skills from resume and JD are processed for semantic similarity (cosine thresholding), powered by Gemini 2.5 Flash LLM.
  • skill_matching_agent.py identifies exact and similar matches between candidate and JD skills.

• Interview Orchestration:

  • planner_agent.py orchestrates the interview in 5 technical questions.
  • question_generator_agent.py dynamically creates questions, tailored to candidate experience and JD requirements.
  • For each candidate answer, context-aware follow-up questions can be triggered.

• Evaluation & Feedback:

  • evaluator_agent.py evaluates candidate answers in real time (hidden scoring stored server-side).
  • Real-time feedback is provided to candidates, with backend maintaining scoring and session details.

• Session Memory:

  • Session data persists using FastAPI backend (extensible to Redis for scalable state management).
  • The memory_agent.py manages conversational and evaluation states.

• Frontend Interface:

  • Interactive chat interface built with Streamlit (streamlit_app.py) for easy candidate interaction.
  • Backend API (orchestrator.py) handles LLM calls, evaluation, and state updates.

Low Resource Usage: Uses only a few LLM calls per interview session (one for generation, up to five for evaluation, plus optional follow-ups), keeping resource use efficient.

---

Suggested Flow Diagram Sections:

• Resume/JD Input → Parsing Agents → Skill Extraction & Matching → Question Planning & Generation → Candidate Answers → Follow-up & Evaluation Agents → Real-time Feedback → Session Memory → Frontend Display[1]

This summary should help map out each module and how data flows from input to candidate interaction and backend evaluation.

---

🚀 Quick Start

1. Clone & Setup

git clone https://github.com/your-org/agentic_recruiter.git
cd agentic_recruiter
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

export OPENAI_API_KEY="your-key-here"
export GEMINI_API_KEY="your-key-here"

uvicorn backend.orchestrator:app --reload

streamlit run frontend/streamlit_app.py

│ └── llm_client.py ├── frontend/ │ └── streamlit_app.py ├── requirements.txt └── README.md