Forwarded from Astronomy Knowledge
Two diamond shaped, rubble-pile asteroids have been observed near Earth, and were photographed by unmanned spacecrafts in 2018 and 2019. Scientists at OIST and Rutgers University have used a simple model normally reserved for the flow of grains to explain their unusual shape. In this image, a photograph of one of the asteroids, Bennu, is shown on the left. On the right, a simulation using the model is shown. As can be seen, the shape of the simulation matches that of Bennu.
I built my own AI News Pipeline (and why?..."readily made" apps weren't enough for me:(
I have seen many apps that claim to deliver customized news, but most of them are hidden systems where you can't control the logic, or they get the facts wrong..i wanted a system that acted as a high level content curator, so I built a custom ETL pipeline to solve this for myself
Here is what you won't find in most AI news apps:
-Parallel Data Architecture: this prevents AI hallucinations by splitting the data stream. the LLM handles the creative rewrite, while the original URLs are preserved in a separate path.
- Local LLM: by running the intelligence layer locally, i eliminated API costs,privacy concerns and third-party subscriptions.
- Smart Ranking & Filtering: delivering only the top 6 highest value stories
As a result,i no longer wake up to a mess of notifications...i just get a professional and summarized briefing of exactly what I need to know
I have seen many apps that claim to deliver customized news, but most of them are hidden systems where you can't control the logic, or they get the facts wrong..i wanted a system that acted as a high level content curator, so I built a custom ETL pipeline to solve this for myself
Here is what you won't find in most AI news apps:
-Parallel Data Architecture: this prevents AI hallucinations by splitting the data stream. the LLM handles the creative rewrite, while the original URLs are preserved in a separate path.
- Local LLM: by running the intelligence layer locally, i eliminated API costs,privacy concerns and third-party subscriptions.
- Smart Ranking & Filtering: delivering only the top 6 highest value stories
As a result,i no longer wake up to a mess of notifications...i just get a professional and summarized briefing of exactly what I need to know
⚡6❤2
Now that i've established the core intelligence and automation engine, i'm exploring ways to refine this into a more robust and shareable solution
Forwarded from Dagmawi Babi
Building AI That Ships
Speaker: Biniyam Daniel
Founder & CEO, AddisAI
Biniyam is the founder of one of the leading AI companies in Ethiopia that has hands on experience from training and fine-tuning their models to building extremely impressive multilingual TTS models.
This talk is a technical talk for every AI enthusiast covering the parts that make AI work. From infrastructure and platforms, to fine-tuning and inference. This will be one of the best and most comprehensive AI talks given in Ethiopia. Excited for you to listen to it.
#DagmawiBabisMeetup #Events
@Dagmawi_Babi
Speaker: Biniyam Daniel
Founder & CEO, AddisAI
Biniyam is the founder of one of the leading AI companies in Ethiopia that has hands on experience from training and fine-tuning their models to building extremely impressive multilingual TTS models.
This talk is a technical talk for every AI enthusiast covering the parts that make AI work. From infrastructure and platforms, to fine-tuning and inference. This will be one of the best and most comprehensive AI talks given in Ethiopia. Excited for you to listen to it.
#DagmawiBabisMeetup #Events
@Dagmawi_Babi
👍1
Florida🛸
photo_2026-01-22_12-03-19.jpg
about the missing eye, i meant to show that what everyone sees on the surface isn't the whole story. if she were to look at you full face, you'd see there is something deeper hidden in her character. it's not necessarily something bad....just those private parts of ourselves that we don't reveal to the world
we'll just call it art🙂
we'll just call it art🙂
❤3🔥1
Forwarded from Chapi Dev Talks
AI / Machine Learning Engineer
Location: Addis Ababa, Ethiopia
Type: Full-time
### About the Role
We are looking for an AI/ML Engineer to join our core research and development team. In this role, you will move beyond high-level APIs to design, implement, and optimize AI models. We are seeking individuals who enjoy the "under the hood" work of machine learning—from data architecture to model convergence.
### Key Responsibilities
* Research and implement state-of-the-art model architectures.
* End-to-end development: Data curation, training, validation, and deployment.
* Optimize models for inference speed and resource efficiency.
* Design experiments to improve model accuracy and robustness.
* Collaborate with the engineering team to integrate models into high-performance environments.
### Technical Requirements
* Advanced proficiency in Python and ML frameworks (PyTorch, TensorFlow, or JAX).
* Strong experience in Model Training (fine-tuning, RLHF, or training from scratch).
* Deep understanding of neural network architectures (Transformers, Diffusion, etc.).
* Experience with high-performance computing and GPU optimization.
### Requirements
* Strong mathematical foundation (Probability, Statistics, Linear Algebra).
* 1+ years of experience in an AI Research or ML Engineering role.
* Ability to think from first principles and solve unconventional problems.
Apply here: https://forms.gle/6LSF4g6vsWt4gYd88
Location: Addis Ababa, Ethiopia
Type: Full-time
### About the Role
We are looking for an AI/ML Engineer to join our core research and development team. In this role, you will move beyond high-level APIs to design, implement, and optimize AI models. We are seeking individuals who enjoy the "under the hood" work of machine learning—from data architecture to model convergence.
### Key Responsibilities
* Research and implement state-of-the-art model architectures.
* End-to-end development: Data curation, training, validation, and deployment.
* Optimize models for inference speed and resource efficiency.
* Design experiments to improve model accuracy and robustness.
* Collaborate with the engineering team to integrate models into high-performance environments.
### Technical Requirements
* Advanced proficiency in Python and ML frameworks (PyTorch, TensorFlow, or JAX).
* Strong experience in Model Training (fine-tuning, RLHF, or training from scratch).
* Deep understanding of neural network architectures (Transformers, Diffusion, etc.).
* Experience with high-performance computing and GPU optimization.
### Requirements
* Strong mathematical foundation (Probability, Statistics, Linear Algebra).
* 1+ years of experience in an AI Research or ML Engineering role.
* Ability to think from first principles and solve unconventional problems.
Apply here: https://forms.gle/6LSF4g6vsWt4gYd88
🔥4
"The way that we think about it is everyone knows about all these LLMs, so ChatGPT and Claude and everything. And they're very good at basically predicting the next token and creating content. But the beauty of agents is you can exploit that feature to help these LLMs kind of navigate a problem on their own. So it's not a chat anymore. You give it a task and you can leave the room. And then this agent's gonna try to autonomously figure out how to get there."
Kieran Flanagan
Kieran Flanagan
🔥3⚡1
#2
When building AI agents, don't ask one model to do everything at once.
breaking down complex tasks (like intent extraction) into smaller, individual components leads to significantly higher accuracy and better overall performance.
1. Accuracy: Smaller tasks reduce the "reasoning gap" for the AI.
2. Efficiency: You can use smaller, faster models (SLMs) instead of expensive LLMs.
3. Scalability: It’s much easier to debug a single component than a massive, complex prompt.
https://arxiv.org/abs/2305.02301
When building AI agents, don't ask one model to do everything at once.
breaking down complex tasks (like intent extraction) into smaller, individual components leads to significantly higher accuracy and better overall performance.
1. Accuracy: Smaller tasks reduce the "reasoning gap" for the AI.
2. Efficiency: You can use smaller, faster models (SLMs) instead of expensive LLMs.
3. Scalability: It’s much easier to debug a single component than a massive, complex prompt.
https://arxiv.org/abs/2305.02301
arXiv.org
Distilling Step-by-Step! Outperforming Larger Language Models with...
Deploying large language models (LLMs) is challenging because they are memory inefficient and compute-intensive for practical applications. In reaction, researchers train smaller task-specific...
⚡2👍2
#3
The concept of Adaptive Prompt Engineering
APE involves dynamically adjusting input prompts based on a model's performance during the training or evaluation phase. instead of "guessing" the best prompt, you use a systematic loop:
- start with initial prompts
- evaluate the output quality
- use techniques like permutation, random search, or LLM based optimization to refine the instructions.
by treating prompt engineering as an optimization problem, you can significantly boost model accuracy and efficiency without needing massive datasets.
APE
The concept of Adaptive Prompt Engineering
APE involves dynamically adjusting input prompts based on a model's performance during the training or evaluation phase. instead of "guessing" the best prompt, you use a systematic loop:
- start with initial prompts
- evaluate the output quality
- use techniques like permutation, random search, or LLM based optimization to refine the instructions.
by treating prompt engineering as an optimization problem, you can significantly boost model accuracy and efficiency without needing massive datasets.
APE
GitHub
GitHub - YiVal/YiVal: Your Automatic Prompt Engineering Assistant for GenAI Applications
Your Automatic Prompt Engineering Assistant for GenAI Applications - YiVal/YiVal
👍2