Tumaini

Hope has your voice

Comment

Inspiration

Umm.. you know, difficult moments are a part of life and I know how difficult it is to keep believing in your goals when everything around you is falling apart. And a lot of times, encouragement from other people may not suffice because they may not fully understand what you are going through. So I thought it'd be nice to hear from the future you who survived all those storms and made it to where you want to be.

What it does

Tumaini is an AI powered emotional wellness tool that helps you hear encouragement from an accomplished version of yourself when going through the lows of life. You tell it about your current struggles, the dreams you're working to achieve despite it and some other personality details and then it clones your voice, creates a personality based off the data you provided (the accomplished version of you) to talk you through the current mess you are in

How I built it

I built this with Nextjs for the client side and fastapi for the backend. I used Gradium for the voice cloning feature.

Challenges I ran into

The biggest challenge I ran in to was getting a free voice cloning api. They were all paid plans that I couldn't afford. I had to do a lot of searching until I found gradium, which wasn't the best but I had to use it because it was free. Then another challenge was getting the ai to generate the message at a conversational pace and not have it sound like a machine. Right now, it's not perfect, but it's a lot more improved.

Accomplishments that I'm proud of

Bringing the project from idea to life is so exhiliarating.

What I learned

I learnt how to use voice models, adjust their stability, fine tune their outputs using stronger prompts, etc.

What's next for Tumaini

I'd like to migrate to a voice cloning ai like elevenlabs which is the best for generating conversational voices and strengthen the pipeline to improve the quality of the generated conversations, allow conversations to reference prior conversations etc all to give the user a robust wellness companion experience.

I'd love for Tumaini to become a full scale mental wellness assistant for people. I want to ship it as a full fledge downloadable application so people can have it straight on their mobile phones and use it.

AI tools used (for best )use of AI track

For the prototyping of Tumaini, I used an agentic ide (opencode with deepseek flash v4 model).

For the voice cloning feature, I used Gradium (an AI voice cloning platform), the free tier is a bit generous and allowed me use it a very limited number of times to clone my voice while building the application. At least it allowed me to show proof of concept before the free credits exhausted (That's why I couldn't add a live link).

For the script and persona generation, I used a free hugging face inference model (Qwen/Qwen2.5-72B-Instruct). It was completely free so I was able to use it to build user persona and generate the script that would be spoken by the voice from the data the user provides.

Built With

Share this project:

Updates

Submission history