Low code data scientist: An end to end approach for code generation for non developers

Low code data scientist: An end to end approach for code generation for non developers


I have been developing this idea for some time now. Here is a big picture view of the approach.The post refers to many of previous posts

  1. The approach uses prompts alone. However, we use conceptual knowledge graphs -which could be complemented by RAG.? (Using conceptual knowledge graph instead of a real knowledge graph with large language models )??
  2. ?The process is based on AI assistance i.e. for a given problem statement,? explore how AI solves them in partnership with a human
  3. The problem is iterative and reflective and is based on Reflection? - including domain based metacognition
  4. The process is primarily designed for Non developers ?and for rapid prototyping but could be used by any developer.?
  5. The metric is the north star i.e. the business metric translated to a machine learning metric is identified upfront and optimised. We used the? jigsaw methodology for teaching AI to non developers. The jigsaw approach could be seen as a larger vision to teach computational thinking /problem solving using large language models such as chatGPT .?
  6. The requirements? analysis ties to the model evaluation metric,? user stories, acceptance tests and hypothesis generation
  7. We use a reasoning engine like OpenAI O1 model to generate Synthetic data by specifying the problem ex regression, the domain (ex healthcare) and the model evaluation metric ex R2?
  8. The code is based on generating code for regression, classification, CNN, MLP and other problems using prompts
  9. Essentially, we are developing a pedagogy and hence it could work with any prompt based tool such as Open AI, cursor, copilot or OpenAI canvas .??
  10. We have been working on a parallel effort for an AI product management?
  11. we were inspired by Magnus (a firefighter from Iceland) ( Magnus Smarason ) and Dr Chougule - an NHS doctor - ( Dr. Amit Chougule DPM,MD,MRCPsych )from our first Oxford course. I continue to collaborate with both of them.?
  12. This work was also developed at the Erdos Research labs - I am a believer in the use of AI for reasoning which is an emphasis for my work.

Vincent Koc

Lecturer, Futurist, and Keynote Speaker | Generative AI Engineer & Technical Leader | Former Top 25 Chief Data & AI Officer | CDAO / CTO

1 个月

Ajit Jaokar im going to cite the visual, great mental model!

Magnus Smarason

Bridging AI Technology with Societal Impact, Making AI Accessible

1 个月

Love the funnel visualization! It’s a fantastic representation of how we can test out different prompts throughout the process, refining and iterating until we get the best results. The structured approach really brings out the power of combining LLMs and low-code AI. Excited to see where this leads!

Rosie Djurovic

Building and innovating | AI | Data | Automation

1 个月

I think for developers, especially juniors, this is also a fantastic approach to accelerate both learning and code production, by having the LLM explain code sections and reasoning behind suggested approaches.

Jaime Canteiro

CFO at Domus Social, EM | MSc student in statistics, mathematics and computing

1 个月

Good thoughts

要查看或添加评论,请登录

社区洞察