Cian Kiplangat

First attempt in developing Kalenjin TTS model that's bit good enough to release it to the public for testing. Laying the foundations of Kalenjin TTS research. Not perfect but a proof of concept that it can work.

@blackie_360 @cursor_ai @ftnabeelah @benln @sunrao Congrats

Huge news I’ve been selected as @cursor_ai Regional Lead for Africa Time to grow the next wave of AI-native builders, startups, and communities across the continent. Africa is building. cursorkenya.com team🫶 Shoutout to @ftnabeelah @benln @sunrao 🙌

@gaagana_ So true, also not forgetting student perks.

Attending university regardless of the program is still necessary, the exposure, opportunities, clubs, events. There's more to it than just the degree. I've benefited from winning a grant. So it's all about how you position yourself I'll say.

@ChatGPTapp It still can make mistakes

@reach_vb I am trying to build a realistic Artemis II mission simulator. With the current technology not everyone can travel to moon and beyond but what if you could get to experience it virtually. x.com/mutaician/stat…

Weekend hack: Build with GPT-5.5 + Codex. Drop your demo in the replies. #1 by likes: 1 year of ChatGPT Pro 2 runner-ups: 6 months each Bonus: Codex picks a wild card winner. Enjoy!

This is what I have now though it doesn't give that wow feeling. I think I overestimated gpt-5.5 + ThreeJS This got me into this rabbit hole of game development because I want to know enough to prompt codex to help me finish the project. 😅

So I ended up getting Codex plus. Couldn't come with something good with the project. For context I want to build a realistic Artemis II mission simulator. With the current technology not everyone can go out of space but what if you could get to experience it virtually.

@mutwiriiii These manual work that I don't want to do but I have to helps a lot. For me huku ndo where ideas just come on it's own

Me hupenda kuosha vyombo. Inafanyanga nikue na a smooth logical thinking process

@kmeanskaran Can one be both?

AI Researcher vs AI Engineering What should you choose? This choice depends on how much you want to read consistently. It is not about tools or scope. Both are goldmine fields. Here is a clean breakdown. AI Researcher: - Very strong grip on mathematics and statistics/probability - Should know mathematical intuitions behind neural networks and Transformers - Focus on innovation and better results on benchmarks - Evaluation is mostly on benchmarks but very competitive - Read papers and implement them - Long-term game that requires patience AI Engineering: - System design is everything - Data preparation and deployment are key required skills - Mostly work on fine-tuning and inference engineering - Focused on multi-agent orchestration - Evaluation metrics are important and often different from on-paper benchmarks - LLMOps, infrastructure design, and cloud skills are crucial to reach the top 10% - Heavy on coding and tools If you are a math ninja and fascinated with the behind-the-scenes of LLMs, then it is suitable for you. If don't have a PhD then working in renowned labs are important. Engineering is my personal favourite. I love building systems and projects that directly affect business metrics. A builder mindset can help you move into product manager roles or launch your own SaaS. Choices are completely personal, but you must pick one side first. Get an income stream and then experiment. Otherwise you will stay in an infinite loop and won't even get a job.

I am in some hackathon having this sort of complex idea. I've depleted Cursor credits and github copilot is not an option now. Now I'm stuck between getting claude code or codex to see if this project can come to live. What do you recommend

@tonykipkemboi Omarchy linux with Brave browser. If I speak it records well but when I play a video and record at the same time it doesn't pick any voice from the video so the recording issue I think its more of how browser's work

@mutaician i'll check on this and fix. what system are you using?

a few weeks ago i created my local version of Wispr Flow. works great and uses the open source models locally with options to use cloud models too. i named it "rekody" (repo in comments). this led me down the rabbit-hole of understanding these Automatic Speech Recognition (ASR) models and how they convert speech to text. i then explored fine-tuning my own version. but then i thought about my mother tongue Kalenjin which has ~6M speakers across Kenya and diaspora. none of the major ASR systems support it. low-resource languages, especially African, are missing almost entirely from the AI models out there right now. if it exists, it's not good at all. several issues that cause this but fixable. so i decided to fine-tune whisper on Kalenjin dataset i found on @huggingface and ran it on @modal. three things I learnt from this experiment: 1. data curation mattered more than model size or training tricks. ~80% of the work was dataset, ~20% was ML. 2. i built it to scratch a heritage itch but didn't anticipate the messages from diaspora Kalenjin friends saying "i understand it but can't speak it. could this become a learning app?" 3. the next step is a Kalenjin text model. so the speech model can eventually help teach the language back to people who lost it. i wrapped the fine-tuned LoRA in an app you can try it out and also a blog write up of the entire build (links in comments).

@tonykipkemboi Tried testing some wav file with less than 60sec of audio but got this error (Content too large). I don't know why my laptop browser doesn't seem to record desktop audios. Seems now I'll do some local tests. As I have some good usecase for this project.

@mutaician that's awesome! would love to check it out. i'm going to start work on text model in Kalenjin.

@_MtuTano Ugali inafaa ikue kwa kisyet

In Kalenjin households, men get served first then the rest😂😂😂equality for who??👀👀

Pitched my Swahili - Kalenjin translator app on some innovation challenge we had at my university (Kabarak) and it was accepted. Hope they deliver what they promised soon 😅 The patience it took me to reach here wasn't easy. This could motivate me to start a startup 😁😂

@ZypherHQ What I really need now is GPU's but I guess the Codex plan can help. I am trying to build models that support my local language. Codex will help me do more research on the algorithms and architectures for training small but efficient models.

This is exactly why I want to do this: If you’re a student building with CODEX/ChatGPT and cost is the only thing slowing you down, I’ll cover 1 month of GPT Plus for 5 builder. Reply with what you’re building. I’ll pick five people in 72h.

@ossia That would be great.

I met with this guy and we're talking about whether freeCodeCamp should develop an AI Assisted Development certification around open source tools and open weight models. What do you think?

@AnkitaxPriya @opencode Which hackathon was this?

perks of participating in hackathons thanks @opencode 🫶

@ClementDelangue interested

Looking for some beta testers for our new robotics dataset hosting features. Who's currently hosting large robotics datasets and would be down to try? 🤖🤖🤖