🤖💡 ContextAI 🤖💡

Author:
Name: Ram Bikkina
Email: [email protected]

---

Project Overview

ContextAI is a Flask-based microservice that takes:

1. A paragraph (context)
2. A sentence (target phrase)

and returns:

• A contextual English explanation of the sentence, based on the paragraph
• A natural, conversational Telugu translation of that explanation

Under the hood, it uses Google’s Gemini-2.0-Flash model (via langchain-google-genai) for both steps. This prototype runs on macOS (M1/M2) with MPS support or CPU fallback.

---

Features

• /explain_translate endpoint (POST)
  • Input: JSON { "paragraph": "...", "sentence": "..." }
  • Output: JSON { "explanation_en": "...", "translation_te": "..." }
• /health endpoint (GET) for quick uptime checks
• Structured logging, error handling, and environment configuration

---

Prerequisites

• macOS (Apple Silicon recommended) or any OS with Python 3.12+
• Git
• Internet access for model downloads
• Google Cloud API Key with access to Gemini models

---

Setup

1. Clone the repo

   git clone https://github.com/ramc26/ContextAI.git
   cd ContextAI

2. Create & activate a virtual environment

   python3.12 -m venv venv
   source venv/bin/activate

3. Install dependencies

   pip install --upgrade pip
   pip install -r requirements.txt

4. Configure environment variables
   Create a file named .env in the project root:

   touch .env

   Add your Google API key:

   GOOGLE_API_KEY= XXXXXXXXXXXXXXXX

   Note: Never commit your real key to Git. Use a placeholder or Git ignore .env.

---

Running the Service

1. Start the Flask server

   python main.py

   By default it will run on http://0.0.0.0:5000/.

2. Health check

   curl http://localhost:5000/health
   # Response: {"status":"ok"}

3. Explain & Translate

   curl -X POST http://localhost:5000/explain_translate \
     -H "Content-Type: application/json" \
     -d '{
           "paragraph": "The weather in Hyderabad today is mostly sunny with a current temperature of 36°C. The forecast for the day indicates that it will remain sunny, with the high also reaching 36°C and the low settling around 26°C. There is no rain expected today, and the wind is a gentle 6 km/h blowing from the southeast. Currently, it feels like 37°C due to the level of humidity in the air.",
           "sentence":  "feels like"
         }'

   Expected response (excerpt):

   {
     "explanation_en": "The sentence \"feels like\" in the context of the paragraph is describing the **perceived temperature** or **apparent temperature**. While the actual measured temperature is 36°C, the level of humidity makes it *feel* hotter, specifically 37°C. This is because high humidity reduces the body's ability to cool itself through sweating, making the air feel warmer than it actually is.",
     "translation_te": "వాతావరణం 36°C ఉన్నా, చెమట వల్ల 37°C లా అనిపిస్తుంది."
   }

---

Sample Test Cases

1. Basic context + sentence

   curl -X POST http://localhost:5000/explain_translate \
     -H "Content-Type: application/json" \
     -d '{
           "paragraph": "Renewable energy sources like solar and wind power are gaining prominence as the world seeks to reduce its reliance on fossil fuels. These sources harness naturally replenishing energy, offering a cleaner alternative for electricity generation. While the initial investment in infrastructure can be substantial, the long-term operational costs are often lower, and they contribute significantly to mitigating climate change. The intermittency of these sources, however, presents a challenge for grid stability.",
           "sentence":  "The intermittency of these sources, however, presents a challenge for grid stability"
         }'

2. Rule-based vs. deep learning context

   curl -X POST http://localhost:5000/explain_translate \
     -H "Content-Type: application/json" \
     -d '{
           "paragraph": "Simple rule-based systems follow if-then rules, while deep learning models learn patterns from data.",
           "sentence":  "The rule-based agent performed predictably under all test cases."
         }'

3. Error handling (missing fields)

   curl -X POST http://localhost:5000/explain_translate \
     -H "Content-Type: application/json" \
     -d '{"paragraph": "Only one field"}'
   # Should return HTTP 400 with {"error":"Invalid JSON payload — 'paragraph' and 'sentence' required"}

---

Notes & Next Steps

• For production deployments, swap out Flask’s built-in server for Gunicorn/Uvicorn.
• Consider caching explanations for repeated queries.
• Extend support to other languages by swapping the second-model prompt.

Enjoy prototyping! 🚀