Inspiration
The project was inspired by the communication gap between customers and service providers in the energy sector. Many customers encounter delays or lack of acknowledgment when they reach out for help or feedback. I wanted to create a lightweight, intelligent system that brings responsiveness, empathy, and data-driven insights to energy support. The goal was to empower both customers and administrators through an accessible, conversational interface that feels human yet efficient.
What it does
Green Wells Energies AI Support Platform is a multilingual Telegram-based assistant that:
- Enables customers to send feedback, ask questions, and chat with an AI support bot.
- Automatically analyzes sentiment (positive, neutral, or negative) using lightweight NLP.
- Categorizes each feedback into service-related areas such as LPG Refills, Pricing, or Complaints.
- Detects and translates messages into English, allowing support for multiple languages.
- Sends daily AI-generated summaries and interactive dashboards to administrators for performance insights.
In short, it turns raw customer messages into actionable intelligence for the Green Wells Energies team.
How we built it
The system was built using Python and the python-telegram-bot framework. The development stack includes:
AI Integration:
- Integrated Google Gemini API for intelligent, contextual responses and daily feedback summaries.
Natural Language Processing (NLP):
- Implemented sentiment analysis with TextBlob and language detection with langdetect.
Multilingual Support: -Leveraged Google Translate API to dynamically translate customer messages and responses.
Data and Analytics:
- Feedback is stored locally in JSON format, categorized, and visualized through a dynamically generated HTML dashboard using Python.
Automation:
- A background scheduler sends daily summaries and dashboards to the admin Telegram account at 7 PM.
Challenges we ran into
- Managing real-time AI response generation while maintaining low latency in the Telegram environment.
- Handling translation edge cases where the detected language or context did not align perfectly with user intent.
- Implementing automatic categorization that works effectively with limited data samples.
- Ensuring feedback data was consistently logged, summarized, and displayed correctly in the analytics dashboard.
- Coordinating multiple asynchronous tasks: feedback handling, AI generation, and scheduled summaries, without blocking the bot’s main loop.
Accomplishments that we're proud of
- Successfully created an end-to-end AI-driven feedback ecosystem that runs entirely within Telegram.
- Implemented multilingual AI support, making the bot accessible to a diverse range of users.
- Automated daily sentiment summaries and generated HTML dashboards for real-time insights.
- Achieved a stable architecture that remains lightweight, modular, and ready for scaling.
- Delivered a system that transforms simple user messages into structured insights for management decisions.
What we learned
This project deepened my understanding of how conversational AI can simplify customer engagement. Key lessons included:
- Integrating APIs like Gemini effectively into asynchronous environments.
- The importance of data labeling and sentiment calibration for accurate results.
- Designing human-AI interaction loops that balance automation with empathy.
- Managing multi-language text streams and feedback analytics in real-world conditions.
- It also reinforced the principle that good AI systems should be transparent, respectful, and adaptive.
What's next for Green Wells Energies AI Support Platform
Next steps include:
- Deploying on the cloud using a scalable backend (e.g., FastAPI + Docker) for continuous uptime.
- Adding a real-time admin dashboard with visual analytics and API integration for branch reporting.
- Enhancing the AI model with fine-tuned feedback data to personalize future responses.
- Implementing speech-to-text and voice support for accessibility in low-literacy regions.
- Integrating customer satisfaction scoring and automatic escalation for unresolved cases.
Log in or sign up for Devpost to join the conversation.