v2

HR Chatbot — v2 (OpenAI Responses API)

A modernized rewrite of the original HR chatbot using an agent loop via OpenAI Responses API directly — no LangChain, no Pinecone. Surfaces real model reasoning via gpt-5.2's reasoning summaries and demonstrates the Responses API's native function calling and file_search built-in tool.

Sample Chat

What changed vs. the 2023 version

	Original (2023)	v2 (2026)
Framework	LangChain 0.0.220	None — pure OpenAI SDK
Model	gpt-3.5-turbo	gpt-5.2
"Thought" panel	LangChain verbose=True scratchpad text	Real reasoning summaries from the model
RAG	Pinecone vector DB + manual embeddings	OpenAI file_search built-in tool
Employee data tool	PythonAstREPLTool (arbitrary code exec)	Typed function tools
Frontend	streamlit-chat (deprecated)	Native st.chat_message / st.status
State	Full message replay each turn	previous_response_id (server-side)

Tech stack

• OpenAI Python SDK >=1.55
• Streamlit >=1.40
• pandas — employee data
• python-dotenv

Setup

1. Create and activate a virtual environment

cd v2
python -m venv .venv

Activate:

• macOS / Linux: source .venv/bin/activate
• Windows (CMD): .venv\Scripts\activate.bat
• Windows (PowerShell): .venv\Scripts\Activate.ps1

2. Install dependencies

pip install -r requirements.txt

---

Pick the path that matches where your LLM is hosted — Path A for OpenAI, Path B for Azure OpenAI.

---

Path A — OpenAI

A1. Configure environment

cp .env.example .env

Fill in OPENAI_API_KEY. Leave OPENAI_VECTOR_STORE_ID blank for now.

A2. Ingest the HR policy (one-time)

python ingest_policy.py

Copy the printed vs_… ID into your .env as OPENAI_VECTOR_STORE_ID.

A3. Run

streamlit run app.py

---

Path B — Azure OpenAI

Requirements:

• A gpt-5.n deployment (or any reasoning model that supports the Responses API)

B1. Configure environment

cp .env.example .env

Set BACKEND=azure and fill in all AZURE_OPENAI_* variables. Leave AZURE_OPENAI_VECTOR_STORE_ID blank for now.

B2. Ingest the HR policy (one-time)

python ingest_policy.py

Copy the printed vs_… ID into your .env as AZURE_OPENAI_VECTOR_STORE_ID.

B3. Run

streamlit run app.py

Files

File	Purpose
app.py	Streamlit UI — renders reasoning panel, tool traces, final answer
agent_loop.py	Responses API streaming loop — dispatches tool calls, yields events
tools.py	Typed function tools + JSON schemas + dispatch table
backend_local.py	OpenAI client + config
backend_azure.py	AzureOpenAI client + config
ingest_policy.py	One-time script to create vector store and upload hr_policy.txt

---

Sample questions to try

• "How many vacation leaves do I have left?"
• "What's the policy on unused vacation leave?"
• "If I encash 10 unused vacation days, how much will I be paid?"
• "Who are the direct reports of Joseph Santos?"
• "Can I apply for sick leave while on probation?"