Nikhil Verma
@nikhilv.bsky.social
CTO at @nonfx_company, previously Ollion @BumbleEng @opera @aol
I love the web and I work to bring functionality to the web people don’t consider
I love the web and I work to bring functionality to the web people don’t consider
Quantum AI! Its Oblivion when you don’t observe it but AI when you do
May 1, 2025 at 3:50 PM
Quantum AI! Its Oblivion when you don’t observe it but AI when you do
NieR Automata is a close second. Japanese studios cook some great music 💓
January 25, 2025 at 6:25 AM
NieR Automata is a close second. Japanese studios cook some great music 💓
I can't use Claude because OpenAI has a zod based guaranteed structured output which it lacks.
January 7, 2025 at 6:21 AM
I can't use Claude because OpenAI has a zod based guaranteed structured output which it lacks.
Bake everything!!!!
December 23, 2024 at 3:01 AM
Bake everything!!!!
are resolved by the few-shot prompting technique.
December 16, 2024 at 7:20 AM
are resolved by the few-shot prompting technique.
It’s funny first i just try to solve the problem with an LLM. Then I make a prompt manually with the first process. Then I ask the LLM to improve its own prompt! “I have this prompt but it gives me this issue pls fix it” and a few iterations later it pretty much perfect. Any issues…
December 16, 2024 at 7:20 AM
It’s funny first i just try to solve the problem with an LLM. Then I make a prompt manually with the first process. Then I ask the LLM to improve its own prompt! “I have this prompt but it gives me this issue pls fix it” and a few iterations later it pretty much perfect. Any issues…
reached it will make the threshold higher and try again. Here is a sample implementation - gist.github.com/NikhilVerma/...
It worked rather well for my usecase
It worked rather well for my usecase
thresholdClusteringWithAdaptiveSimilarity
thresholdClusteringWithAdaptiveSimilarity. GitHub Gist: instantly share code, notes, and snippets.
gist.github.com
December 13, 2024 at 6:28 AM
reached it will make the threshold higher and try again. Here is a sample implementation - gist.github.com/NikhilVerma/...
It worked rather well for my usecase
It worked rather well for my usecase
Yeah I had a similar issue where I don't want to give the total cluster count. Kmeans has the elbow method to detect ideal cluster size. But what worked well for me was a threshold + max cluster size approach. The algorithm works by clustering with a specific threshold, however if max size is...
December 13, 2024 at 6:28 AM
Yeah I had a similar issue where I don't want to give the total cluster count. Kmeans has the elbow method to detect ideal cluster size. But what worked well for me was a threshold + max cluster size approach. The algorithm works by clustering with a specific threshold, however if max size is...