Required stack
필요 기술
PythonNode.jsNatural Language ProcessingAI ChatbotConversational AIMobile App DevelopmentWeb DevelopmentSpeech-to-TextText-to-SpeechDatabaseCloud API IntegrationSpaced Repetition System (SRS)
Project brief
프로젝트 내용
I want to launch a conversational practice bot that feels like a friendly language partner rather than a rigid tutor. The core flow starts with the bot itself proposing everyday scenarios—ordering a cup of chai at a Connaught Place café, asking an auto-rickshaw driver for directions, checking in at a hotel in Jaipur—and guiding the learner through the dialogue. As we talk, the system must recognise and respond to natural Hinglish, yet steer the learner gently toward polished, formal Hindi without sounding patronising.
Right after each session, the same chat history should feed a dynamic quiz: the bot plucks out the trickier words or phrases (for example anubhav, prakriti), schedules them with spaced-repetition logic, and surfaces them in later chats until the learner shows mastery. No static vocabulary lists—everything adapts to the user’s real mistakes and hesitations.
Target environments are both a mobile app and a web interface. Learners will be free to type or speak, so the build needs seamless text I/O plus speech-to-text and text-to-speech hooks. I am open to whichever stack you are comfortable with—Rasa, Dialogflow CX, or a custom NLP pipeline in Python or Node.js—so long as:
• the scenario engine can expand easily with new role-plays,
• Hinglish code-switch detection is accurate enough to nudge, not scold,
• quizzes follow a true SRS algorithm (SM-2 or similar), and
• voice and text remain in sync across web and mobile.
Please include a brief outline of your proposed architecture, any pretrained language models or Hindi ASR/TTS services you would leverage, and a timeline for an MVP that covers at least three role-play scenes and the adaptive quiz loop.
Right after each session, the same chat history should feed a dynamic quiz: the bot plucks out the trickier words or phrases (for example anubhav, prakriti), schedules them with spaced-repetition logic, and surfaces them in later chats until the learner shows mastery. No static vocabulary lists—everything adapts to the user’s real mistakes and hesitations.
Target environments are both a mobile app and a web interface. Learners will be free to type or speak, so the build needs seamless text I/O plus speech-to-text and text-to-speech hooks. I am open to whichever stack you are comfortable with—Rasa, Dialogflow CX, or a custom NLP pipeline in Python or Node.js—so long as:
• the scenario engine can expand easily with new role-plays,
• Hinglish code-switch detection is accurate enough to nudge, not scold,
• quizzes follow a true SRS algorithm (SM-2 or similar), and
• voice and text remain in sync across web and mobile.
Please include a brief outline of your proposed architecture, any pretrained language models or Hindi ASR/TTS services you would leverage, and a timeline for an MVP that covers at least three role-play scenes and the adaptive quiz loop.