Ali Cihan Isikal

Tangy AI

Summary

Putting Gen-AI concepts of RAG and custom embeddings based on user’s documents to generate related learning content, mainly flashcards and tests helping users a diverse learning experience with feedback on their test results.

Key Points

• I feel kind of fascinated how some people are really quick to pick up new stuff. At the same time i was frustrated with learning topics that i have no motivation to learn. For example material science class in uni and physics wasn’t my tea, and even though you study those topics feedback loop on learning always takes more time, making learning suck more ass on the run. So i wanted to include an explanation part to the uncorrect answers, so learner can try to understand by themselves to get to the right solution.

• Initial thought was to use RAG to generate questions and flashcards solely based on given info. For RAG to work we needed to generate custom embeddings from the document.

• Uploading PDF file, taking the text content, sending over an API for generating embeddings, using embeddings to context to generate flashcards/tests.

• I really liked playing with prompts, some prompt engineering helped me to get the right outcomes.

• I also integrated RevenueCat as my in-app purchase processor, it took some time, but i rather this than writing server-side code.

• Can use more energy and thoughts on this project but consumer projects are somewhat had this mirage.

• Future improvements:

  • Making services run faster, generating test and flashcard responses takes some time that can bother user.
  • Spend time with social-media content creation to boost growth of user base with an ai-generated influencer.