From Engineering to Biology: Why You Should Master AI

From Engineering to Biology:
Why You Should Master AI

aleksandr.avdiushenko@jetbrains.com

These slides at avalur.github.io

Everybody Has a Brilliant AI Assistant – It Always Ready to Help!

✨ Guide your thinking process like a pro
🔍 Quickly search for the most relevant information
💡 Effortlessly generate insights that spark ideas

We can focus on what truly matters: critical thinking, decision-making and verification

Examples: Five current projects by students from our CSAI program

Tic-tac-toe with Robot Arm
Animations in Manim for Solving Geometry Problems
Cryogenic Electron Microscopy
LLM for Generating Astronomy Olympiad Problems
Competition on Boolean Circuit Synthesis: 2024 → 2025

1. Tic-tac-toe with Robot Arm

General idea
- Implement game with robot arm
Type
- Engineering / Research / Fun
Equipment
- Kinova robot arm
- Raspberry Pi / Optitrack cameras
Outcome
- Robot wins human

CSAI NUP. Kirill Krinkin. Spring 2025.

AI in Robotics

Eureka: Human-Level Reward Design via Coding LLMs (ICLR 2024)

Further after the first project I would like to share a groundbreaking paper that was recently presented by Nvidia at top-level conference ICLR 2024. It is titled 'Eureka: Human-Level Reward Design via Coding Large Language Models.' In my opinion, the main feature of this paper is that it uses Large Language Models (LLMs) to design human-level reward functions. As you see, the task is really hard — a robot hand should be able to balance and rotate the pen. It is a challenging task even for humans, and probably it is even more challenging for humans to explain how our muscles work in this case. But Coding LLMs can help us solve the problem of designing reward function for this hard task. And eventually the robot hand is able to balance and rotate the pen as some humans do!

2. Animations in Manim for Solving Geometry Problems

Develop specialized geometric tools to enhance Manim's capabilities for geometry problems
Create AI system to automatically generate Manim code from problem descriptions
Build converter to transform GeoGebra constructions into Manim animations

CSAI NUP. Fedor Baharev. Spring 2025.

Let's move on to the second project, which combines mathematical visualization with AI in a really creative way. Here students are working with Manim, the same animation engine used by the popular maths YouTube channel 3Blue1Brown. As you can see from this visualization, Manim is capable of creating really nice mathematical animations. This project has three ambitious goals: First, students are extending Manim's toolkit by developing specialized geometric tools. The second is building an intelligent system that can take a written geometry problem and automatically generate the Manim code needed to animate it. And third, students will build a bridge between GeoGebra and Manim by developing a converter. This will allow us to take existing geometric constructions from GeoGebra and transform them into dynamic Manim animations. The teamleader here is Fedor Baharev, one of our professors.

3. Cryogenic Electron Microscopy

Machine Learning in Classification: ML-driven 2D/3D classification using maximum likelihood optimization and PCA for conformational analysis
Intelligent Particle Detection: CNNs revolutionize automated particle recognition, enhancing accuracy over traditional methods
Advanced Model Generation: Modern deep learning approaches using VAEs and GANs enable robust initial 3D structure determination

CSAI NUP. Fedor Baharev. Spring 2025.

Semantic segmentation-based detection algorithm for challenging Cryo-EM RNP samples

4. LLM for Generating Astronomy Olympiad Problems

Current Challenge: Existing LLMs fall short in generating high-quality olympiad problems that match Iternational Olympiad standards
Research Objectives: Develop specialized datasets and explore novel approaches for training AI models to generate problems

CSAI NUP. Alex Avdiushenko. Spring 2025.

The next project addresses a challenge in educational AI: using LLMs to generate high-quality astronomy olympiad problems. Let me highlight the two main aspects of this project:
First, we're addressing a crucial limitation in current technology. Despite their sophistication, existing LLMs struggle to generate astronomy problems that truly meet International Olympiad standards. These problems require not just factual knowledge, but deep understanding of astronomical concepts, mathematical reasoning, and the ability to create engaging, challenging scenarios.
Second, we're focusing on two key research goals:
— Developing specialized, carefully gathered datasets that capture the essence of high-quality olympiad problems
— Exploring novel training approaches to help AI models understand and replicate the complex reasoning required for these problems

5. Competition on Boolean Circuit Synthesis:
2024 → 2025

CSAI NUP. Alexander S. Kulikov. Spring 2025.

Five current projects by students from our CSAI program

Tic-tac-toe with Robot Arm
Animations in Manim for Solving Geometry Problems
Cryogenic Electron Microscopy
LLM for Generating Astronomy Olympiad Problems
Competition on Boolean Circuit Synthesis: 2024 → 2025