I recently spent some time understanding what “dimension” actually means when we talk about embeddings.
We often hear: 512, 1024, 1536 dimensions...
But what exactly is a dimension?
Read Here: https://t.co/cQwgpvZ09K
#rag#agenticai#generativeai#ai
India mein koi 5 minutes number Google kar le to doodh ka doodh paani ka paani ho jaye.
RBI surplus to Govt : ₹2.87 Lakh Crores (US$30 Bn)
Total listed bank profits : ₹4.11 Lakh Crores (US$42 Bn)
NPCI - which runs UPI pre tax ‘surplus’ : ₹ 1,888 crores (US$200 Mn)
To nuksaan kis ka ho raha hai UPI se aur kaunsi subsidy de rahi hai Govt UPI pe jo chubh rahi hai?
Cost of running ATMs and cash logistics in India is ₹30,500 crores (US$ 3.0 Bn). If you want to optimise shut ATMs and promote UPI instead.
Any levy on UPI is just tax collection. UPI is the one scientific achievement of India everyone acknowledges par ab tax ki bali chadhegi.
One thing I found interesting while working on this:
The biggest improvement doesn't always come from changing the LLM.
better question is:
Are we even giving the LLM the right information?
Good retrieval + good ranking can be just as important as the model.
#RAG#LLM#AI
Recently I learned that RAG is much more than:
Retrieve documents -> Send to LLM -> Get answer.
That approach can work for a demo.
But in prod, one of the hardest problems is simply getting the right information in the first place.
made me look deeper into retrieval. #RAG#AI
Building the retrieval pipeline is only half the work.
We also need to know: “Is this actually better?”
I worked with benchmark datasets and retrieval metrics to compare different configurations.
AI systems need evaluation, not just implementation. #AI#RAG
Even after hybrid retrieval + RRF, the results are not always in the perfect order.
the next step is reranking.
A Cross-Encoder can look at the query and retrieved content together and score how relevant they are.
Dense + Sparse - RRF - Re-ranking - LLM
#RAG#AIEngineering
When we have results from both dense and sparse retrieval, we need a way to combine their rankings.
That's where Reciprocal Rank Fusion (RRF) comes in.
It combines rankings from different retrieval methods.
Small change, but big difference in retrieval quality. #RAG#Search
How we can improve retrieval quality.
One approach is using both:
• Dense vectors → understand meaning
• Sparse vectors → match important keywords
We can combine both to get better results.
This is where hybrid search becomes interesting. #RAG#AI
After my last learning update, I explored NumPy, Pandas and Pydantic.
Didn't go too deep yet. Just focused on the concepts I think I'll need for now.
I'll go deeper when I need them in projects.
Continuing my Python → AI learning journey. #Python#AI
Starting my AI Engineering journey - one step at a time
I decided to start going deeper into AI Engineering.
I know that if I want to work seriously in AI, I first need to be comfortable with Python.
So, mostly using my weekends and free time, I started learning Python.
Met the Hon’ble Lok Sabha Speaker today along with MPs of the Opposition.
Our demand is simple: Parliament must have a detailed discussion on the brutality unleashed on students yesterday and on the government’s complete lack of accountability for the examination crisis.
Students were beaten for asking legitimate questions about their own future.
If Parliament cannot discuss the future of India’s youth, what is it for?
The Opposition will not let this be buried. We will ensure that the students’ voice will be heard on the streets, and in Parliament.
Thinking of working on a political-tech idea. The opportunity feels huge, but the current political environment makes me a little nervous.
Would you build in this space?
This weekend, I updated my portfolio design.
The previous version was decent, but it didn’t feel right. So I spent time refining the layout, improving clarity, and making it more polished overall.
Visit: https://t.co/1HUqEsqCLF
@ShainaRoyyy It's really very hard to find something which is not built yet.
Try to get rid of that fear people can copy your idea not your execution and passion.