Hacker News: Pranav_Ghoghari

New comment by Pranav_Ghoghari in "Special Projects (2016)"

Pranav_Ghoghari — Fri, 25 Sep 2026 09:59:56 +0000

Its really impressive that they stuck to their vision over the long term. This is difficult and more challenging than it sounds.

New comment by Pranav_Ghoghari in "Introducing System One Models and Jev"

Pranav_Ghoghari — Sun, 20 Sep 2026 10:59:58 +0000

I am not sure for how long the output will stay absolutely free. But apart from the pricing advantage of Jev itself, I just love the simplicity of having only an input price. Input is pretty easy to estimate and calculate upfront, which makes the cost of running something at scale much more predictable.

With LLMs, even with JSON schema constraints and structured output, the actual cost can still be hard to predict because of varying output lengths and, especially, unpredictable reasoning costs. There is something really nice about being able to tokenize and predicting the budget beforehand.

New comment by Pranav_Ghoghari in "Ask HN: What are you working on? (September 2026)"

Pranav_Ghoghari — Wed, 16 Sep 2026 13:46:42 +0000

Trying to take Vector DBs beyond RAG.

Thinking about making an open-source library specific to Vector related operations like clustering, classification, anomaly detection to make it easy for developers to get more out of their vector embeddings. Basically sk-learn for Vector operations.

New comment by Pranav_Ghoghari in "Ask HN: If embeddings are so powerful, why are they mostly used for retrieval?"

Pranav_Ghoghari — Sun, 13 Sep 2026 17:15:10 +0000

Trueee,,,the use cases normally are very engineering oriented and not useful for every end user.

Ask HN: If embeddings are so powerful, why are they mostly used for retrieval?

Pranav_Ghoghari — Sun, 13 Sep 2026 16:38:42 +0000

Vector embeddings natively are capable of much more than just retrieval. Even when OpenAI released their initial embedding models, it mentioned that their embeddings are natively capable of Search, Clustering, Recommendations, Anomaly Detection, Diversity Measurement, and Classification.

But everything is concentrated around Vector DBs and RAG, and is progressing towards GraphRAG and Knowledge Graphs. Why are the other representations being underused? Like, why didn’t they catch up?

Everyone is complaining that RAG is dead, Vector DBs are done, they are expensive, and the ROI is not worth it. I know most of this is noise. Embeddings are meaningful and a part of many production pipelines. But why don’t these people try to get more value out of their existing stored embeddings?

I don’t know if it’s just me being delusional, but I feel there is a lot of hidden potential that is untapped. Why is retrieval so dominant? I mean, yeah, search has throughout history been a really important operation across any data. But the purpose of embeddings itself was to capture semantic relations beyond just search. Is it because search is the easiest to productize and build features around? And other use cases are not so frequent?

Comments URL: https://news.ycombinator.com/item?id=49685838

Points: 2

# Comments: 2