today was a long day; two events occupied it: the elevenlabs hackathon and the skit for the mission conference at IEC SF.

highlights at the elevenlabs hacks:

• free t-shirts were good quality with heavyweight cotton and embroided logo
• nice stickers
• free breakfast
• free lunch that was not pizza
• nice office with drinks and snacks and coffee machine

below was the prompt i used for lovable to get started. i don't like how i can't even read the code without pushing to github.

here’s the new flow path for the app start with a record buton in the mdidle (R) click play to record and click pause to stop and we have two butons, one on left and one on right, left is query mode (Q) and right is guide mode (G) left button is query mode (Q), which takes my transcription, and passes it to LLM, it answers my question absed on the image, and generates text, and elevenlaabs answers me. here there should be three buttons, (Q, R, G) and after answering, i can record to ask another question, or click the back buton to go back to the home screen of three buttons (Q, R, G) when i click guide mode, it should ONLY speak to me when there’s anything potentially dangerous or things i should watch out for. but other than that, it should not speak. it should be taking screenshots every 5 seconds here. and theres hould be two butons here. a back buton on the left, and record button on the right to pause the screenshot and ask what’s going on

highlights at mission conference

• skit by SJ was interesting
• skit was fun to perform, A watched us
• met a pastor who wants to evangalize to other parts of indonesia with their own type of javanese, and utilize llms for that
• good indonesian food

went back after church and decided to stay until 10 p.m. to code speech-to-text. first we started with assemblyai, but streaming was pay-to-use. and it took 30 minutes to even figure that out since. next to try was deepgram, but given the time constraints, and my thinning patience, we went for openai to transcribe for us. turns out you ned to create a file in some temp directory after recording, and then pass it as a certain audio file to openai.

went back home to continue integrating and merging all our work. we ended up staying up till 6 a.m. because of cursor and claude generating a bunch of code that made it hard to debug and work with each other.