🤝 SmartColleague

SmartColleague is an AI-powered assistant designed to improve how employees and clients interact with a company's internal knowledge — from HR processes and employee expertise to documentation and project history.

---

👩‍💼 Contributor

• Maria Rosaria Di Domenico — Data Scientist

---

🧩 Problem

Companies often struggle with scattered internal knowledge:

• Employees are unaware of who's working on what.
• Documentation is hard to navigate.
• HR procedures are opaque or difficult to follow.

This results in reduced productivity, onboarding friction, and internal communication gaps.

---

🚀 Introducing SmartColleague

SmartColleague is your intelligent workplace assistant — designed to help employees, HR teams, and even prospective clients access and understand company information effortlessly.

It enables users to:

1. 🔍 Discover people, projects, or documents using natural language.
2. 🧭 Navigate HR processes with clear, step-by-step assistance.
3. 📄 Access company documents or find the right expert for a task.
4. 📎 Analyze a PDF file uploaded on-the-fly, not included in the company database.
5. 💬 Interact with a fallback general-purpose AI model to ask off-topic or general knowledge questions.

🎯 Goal

To build an intuitive and helpful assistant with access to different facets of company knowledge — enabling better collaboration, document discovery, HR support, and personalized experiences.

---

🧠 Technical Stack & Architecture

SmartColleague integrates advanced Generative AI and modern NLP technologies:

• LLM Core: Google Gemini (gemini-1.5-flash) for understanding queries, generating responses, and calling tools.
• Agent Framework: LangGraph to manage conversational state and tool orchestration.
• Key Tools & Capabilities:
  • Vector Database (ChromaDB): For semantic document and resume search.
  • Document Parsing: PDF processing to read and extract information from resumes, policies, and company docs.
  • Embedding Models: For chunking and indexing documents.
  • Function Calling / Routing: Identify whether the user wants HR help, document retrieval, team discovery, file analysis, or general QA.
  • Fallback LLM Support: For general-purpose non-corporate queries (e.g., similar to ChatGPT).

---

📚 Data Sources

SmartColleague relies on several internal data sources:

1. Resumes / CVs: Parsed from PDF files (~1K+ examples).
2. HR Documentation: Policies, procedures, benefits guides, etc.
3. Company Docs & Projects: Internal knowledge bases, project reports, presentations.

---

💻 Setup & Installation

Follow these steps to run SmartColleague locally:

1. Clone the repository:

git clone https://github.com/your-username/smartcolleague.git
cd smartcolleague

2. Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

Or manually install key libraries:

pip install google-generativeai langchain-google-genai langgraph chromadb pypdf openai

4. Add API keys:

Store your Google Gemini API key securely using .env:

GOOGLE_API_KEY="your-api-key-here"

Or set it directly via environment variables.

---

✅ Current Capabilities

1. 📄 Retrieve resume or document content based on semantic search.
2. 🧑‍💼 Match employee profiles to a specific user query.
3. 📂 List relevant documents or projects by topic.
4. 📘 Provide HR policy and procedure assistance.
5. 📎 Analyze newly uploaded PDFs (even if not part of the preloaded database).
6. 💬 Answer general, non-corporate questions using a fallback LLM (ChatGPT-style).
7. 🧠 Understand context and conversation history via chat interface.
8. 💾 Summarize or extract insights from project files or onboarding docs.

---

🧪 Example Use Cases

• “I need someone with NLP experience for a client project.”
• “Where is the latest performance review process document?”
• “Analyze this PDF I just uploaded — what’s the key takeaway?”
• “What is the difference between supervised and unsupervised learning?”

---

🎨 UI/UX

SmartColleague is deployed with a Gradio interface, enabling users to:

• Ask questions like:
  “Who in the company has experience with AI projects?”
  “Show me the onboarding guide for interns.”
• Upload files for quick summarization or extraction.
• Ask general-purpose questions unrelated to the company (e.g., “What’s the capital of Norway?”).

---

📌 Notes

• Ensure your API keys are never hardcoded in public repositories.
• For production deployment, consider containerizing with Docker and using secure storage for secrets.

---

📬 Feedback & Contributions

We welcome feature suggestions, bug reports, and contributions. Feel free to fork this repo and open a pull request!