QueryBot
QueryBot – Your Real-Time Multilingual AI Tutor. Empowering students with instant voice, and text-based answers across languages—anytime, anywhere.
Created on 20th April 2025
•
QueryBot
QueryBot – Your Real-Time Multilingual AI Tutor. Empowering students with instant voice, and text-based answers across languages—anytime, anywhere.
The problem QueryBot solves
QueryBot addresses a crucial gap in modern education by providing accessible, real-time, multilingual academic support to learners across the globe. In today’s digital learning environment, millions of students struggle with accessing quality academic assistance—especially those in rural or underserved areas. Traditional tutoring is either unaffordable or unavailable, and students are often left without help when they need it the most, especially during odd hours or in non-school settings. QueryBot solves this by offering an AI-powered tutor that is available 24/7, capable of understanding voice, text, and image inputs, and responding instantly in multiple languages.
Another major barrier in education is language. Many academic resources are available only in English, leaving students who are not proficient in it at a disadvantage. These students often find it challenging to comprehend complex concepts due to language barriers. QueryBot removes this obstacle by offering multilingual responses, enabling students to receive explanations in their native languages. This inclusivity helps democratize learning by ensuring that students no longer have to rely solely on English resources.
In addition to language barriers, students often experience cognitive overload due to the fragmented nature of educational tools. They switch between apps for searching questions, translating content, solving image-based problems, and interacting with tutors. This fragmentation results in wasted time, reduced focus, and increased frustration. QueryBot consolidates all of these functions into a single seamless platform, allowing users to speak or type questions, upload images of doubts, and receive meaningful, contextual answers—all in one place. This unified experience improves productivity and ensures a smoother learning process.
Accessibility is another key concern that QueryBot tackles. Differently-abled learners, such as visually impaired students or those with limited motor skills,
Challenges I ran into
While building QueryBot, one of the most significant challenges was integrating multimodal inputs—specifically handling image, voice, and text data simultaneously in a seamless user experience. Coordinating these inputs in real time within a React frontend and sending them properly encoded to the backend API required in-depth debugging, especially while dealing with multipart/form-data requests. Inconsistent browser behavior across devices made it even harder to ensure voice recording and image upload features worked reliably everywhere.
Another major hurdle was dealing with language support and multilingual responses. While the language model performed well in English, generating accurate and contextually appropriate responses in regional languages like Hindi and Bengali was initially inconsistent. We solved this by explicitly setting the language context in the prompts sent to the LLM (e.g., “Answer this in Hindi like a teacher talking to a 10th-grade student”), which dramatically improved the quality of non-English responses.
We also faced a tricky CORS issue when making cross-origin requests between the React frontend hosted on Vercel and the backend API running Flask on Render. After some trial and error, we configured proper CORS headers using flask-cors, tested various deployment configurations, and ensured that all endpoints were exposed securely but accessibly.
One particularly frustrating bug was the voice input not working properly on some Android devices. The speech recognition library behaved differently across platforms, and we had to switch from the default Web Speech API to a custom library with fallback support. Extensive mobile testing helped us identify the exact scenarios where the feature broke, and we patched it accordingly.
Lastly, we had a challenge with optimizing the response time of the app. Since the LLM queries (especially for image processing) could sometimes take a few seconds, we added loading states, audio feedback cues, and impr
Tracks Applied (1)
Groq track
Groq