⬛️ tip: llm-chaining (easy and powerful)

here’s a really short tip i wanted to share (ahead of next week’s much much longer post)

llm-chaining

most people prompt one llm and stop. you can get great results come from chaining multiple lllms together.

here's an animated visualisation of what i mean (let it load) 👇

in this example (this was for email reply automation as well as research)

→ start with chatgpt 3o for structure + ideation
→ send the draft to perplexity for fast fact-checking
→ pass the refined output to claude 3 for synthesis + tone
→ run the synthesis through r1 for polish
→ finish with a final check and polish in chatgpt 4.5 before shipping

each model has strengths. chaining lets you stack them.

(don't ask me why i used r1 in the end, i tinkered and it just worked better that way.)

bonus tip
step 1: start manual = open a bunch of tabs and do this manually (or use something like lmstudio)

step 2: automate using something like lleverage, relevance or n8n

thx for sharing this tip with me Walid Boulanouar

⬛️

other posts you might have missed