🚀 AI Q&A Bot

A simple AI-powered Question Answering Bot, built as part of my intern assignment. It started as a command-line app and later evolved into a Streamlit web app, deployed on Hugging Face Spaces.

---

📌 Why I Chose This Project

I chose AI Q&A Bot because I’ve always been interested in having an AI of my own for gathering and summarizing information. Initially, I wanted to connect it to the internet for real-time answers, but for this assignment, I focused on building the foundation step by step.

---

🛠️ Development Journey

Step 1: First Run Attempt

• I started with OpenAI’s openai.ChatCompletion.create().
• Error I got:

openai.lib._old_api.APIRemovedInV1:
You tried to access openai.ChatCompletion, but this is no longer supported in openai>=1.0.0

Fix: Migrated to the new SDK (client.chat.completions.create) instead of downgrading.

---

Step 2: Fixing ImportError

• Error with from openai import OpenAI.
• Cause: My OpenAI library version was mismatched.
• Solution: Upgraded to OpenAI v2.0.1, reinstalled, and confirmed import worked.

---

Step 3: Quota Issue

• Eventually hit this error:

openai.RateLimitError: You exceeded your current quota

• My OpenAI free credits expired.
• Solution: Instead of adding billing, I switched to Hugging Face models that run locally for free.

---

Step 4: Switching to Hugging Face

• Model chosen: google/flan-t5-small

• Why?

  • Lightweight (~80MB).
  • Runs on CPU (no GPU needed).
  • Can handle Q&A, summarization, and general instructions.

---

Step 5: Transformer Import Error

• When using Hugging Face pipeline, I got:

ModuleNotFoundError: Could not import module 'pipeline'
RuntimeError: operator torchvision::nms does not exist

Fix:

• Uninstalled torchvision (not needed for text models).
• Installed CPU-only torch.
• Used AutoTokenizer + AutoModelForSeq2SeqLM instead of pipeline.

---

Step 6: Minimum Working Bot

Finally, I got the bot running locally! 🎉

Example Run

🤖 Welcome to the Local AI Q&A Bot (Flan-T5 Small)!
Type 'exit' to quit.

You: tell me about yourself
Bot: if you are a sexy person, you may want to know about your sexy personality.

Reflection

• The bot works, but small models sometimes give funny/off-topic answers.
• The assignment goal was effort, resourcefulness, and creativity — and I achieved that.

---

Step 7: Adding a Streamlit UI

To go beyond the minimum, I built a simple web interface using Streamlit.

How to Run Locally

pip install streamlit
streamlit run app_streamlit.py

Screenshot

Reflection

• Made the bot more polished and user-friendly.
• Even though the model is small, the UI makes it easier to interact with.

---

Step 8: Deployment on Hugging Face Spaces

To make the project accessible anywhere, I deployed it for free on Hugging Face Spaces.

Deployment Steps

1. Created a Hugging Face account.
2. Created a Space (Streamlit SDK).
3. Uploaded app_streamlit.py, requirements.txt, and README.md.
4. Hugging Face auto-built and launched the app.

Result 👉 AI Q&A Bot on Hugging Face Spaces

---

⚡ Future Improvements

• Add conversation history in the UI.
• Allow choosing between multiple models (flan-t5-small, distilbert, etc.).
• Connect to the internet for real-time answers.
• Fine-tune or train the model with reinforcement learning (RL) to improve accuracy.