Agents Save the day

Building low cost scalable agents, industry uses 100b parameter models, we will try to achive the better results using smaller models 7-b, by fine-tuning slms . easily scalable and low cost.

Comment

goal is to build fast, cheap and scalable agentic ai which analyses timeseries f1 data in seconds. methods pret-train and fine-tune SLMs 7b-8b to analyse f1 data.

example Lewis Hamilton asks: “Give me a detailed review of tire temperature throughout the race and compare with the USA one.” An AI system with agents works like this: Agent 1 gathers recent race data from the 2025 Monaco Grand Prix. Agent 2 retrieves tire temperature data for both 2025 USA GP and 2025 Monaco GPs. Agent 3 compares the two datasets, highlighting how tire temperatures evolved across races. Finally, Hamilton gets a concise report explaining how his tire temperatures behaved this race versus the previous year with the option to choose any past race for comparison.

How we optimise costs:

GPT-4o mini $0.15/million input tokens, $0.6/million output tokens Per complex prompt (with image and detailed analysis): $0.0007

GPT-o1-Pro. $150 / million input tokens $600 / million output tokens Per complex prompt (with image and detailed analysis):$0.66 per complex prompt

if we use 5 agents of gpt-4o mini we spend 0.0035$!!👌🚀 if we use 5 agents of gpt-o1-pro mini we spend 3.3$!!😭😭

That is 942 times cheaper ✌️ and with gpt-4o mini you will get better results and models are fine-tuned and trained for specific results.

RL comes to save letss gooo!!

Built With

  • langgraph
  • openpipe
  • pocketflow
  • slm
  • unsloth
Share this project:

Updates