Tarun Kumar Singh

AI/ML Engineer (Python + Django) | LLM/RAG Architect, Agents, Forecasting, MLOps | AWS/GCP | OpenAI | Vector DBs (pgvector/FAISS) | I ship AI SaaS with clean UI

[Tarun Kumar Singh @krtarunsingh]

@BaT_MaN_23 @grok @grok Create an image of a classy mini-storey apartment building designed for long-term rental income. Make it modern, stylish, attractive, and convenient, with well-planned apartments that tenants would love to live in.

[Tarun Kumar Singh @krtarunsingh]

GPT-6 Astra Can Actually Use Your Computer 🤯

GPT-6 Astra isn’t just better at answering questions — it can actually take action on your computer.
OpenAI’s newest model can browse websites, use software, write and test code, analyze documents, and create spreadsheets and presentations while handling multi-step tasks.
It’s also OpenAI’s first model to reach the Critical cybersecurity capability threshold.
ChatGPT is starting to look a lot less like a chatbot… and a lot more like an AI worker.
#gpt6 #GPT6Astra #chatgpt #ai #openai

[Tarun Kumar Singh @krtarunsingh]

@TeamYouTube Hey i followed, how can we continue our conversation in DM? @TeamYouTube

[Tarun Kumar Singh @krtarunsingh]

Hi @TeamYouTube, I need help with a terminated channel that appears to have been hijacked.

Channel: https://t.co/bX48eNrWIL

I used the Hacked Channel Assistant, secured my Google Account, and it directed me to the termination appeal form. However, that form will not open a new review because a previous appeal was recently denied.

I recently reviewed my Google Takeout and found multiple uploads I do not recognize, including crack/hack/view-bot videos. The videos also contain unrelated branding such as “PABLO INC.” which has never been associated with me or my channel.

I previously contacted YouTube regarding possible hijacking under cases 7-8755000036914 and 8-0716000037700.

Could you please help me access the appropriate hijacked-channel review process or escalate this to the hijacking team?

[Tarun Kumar Singh @krtarunsingh]

@TeamYouTube Done Follwoing

[Tarun Kumar Singh @krtarunsingh]

Claude Fable 5 is back — and this feels bigger than just another AI model update.

Its return shows where frontier AI is heading: stronger reasoning, better coding support, deeper agentic workflows, and more serious conversations.

Read here: https://t.co/YiAeg3mqse

[Tarun Kumar Singh @krtarunsingh]

I tested Claude AI as a developer and compared it with ChatGPT for real writing, coding, planning, and document-heavy work.

Read the full article here: https://t.co/s3vvHOD3Sn

#AI #ClaudeAI #ChatGPT #Python #Developers #ArtificialIntelligence #TechWriting

[Tarun Kumar Singh @krtarunsingh]

Just read this — solid breakdown of a real-world problem with a practical mindset.

Loved how it focuses on execution over theory. This is the kind of thinking more builders need 👇

Visit: https://t.co/8VsaQqUPt3

#AI #Tech #Development #Learning

[Tarun Kumar Singh @krtarunsingh]

Most RAG apps aren’t improving.

They’re just doing vibe testing.

I built a Python evaluation blueprint to measure:
→ retrieval quality
→ answer accuracy
→ latency
→ cost
No more guessing.

Full guide: https://t.co/OUIfYiu8fG

#AI #RAG #Python #LLM
cc: @OpenAI @huggingface @LangChain @llama_index

[Tarun Kumar Singh @krtarunsingh]

Most RAG systems don’t fail because of the LLM.
They fail because of bad retrieval.
When I first built a RAG pipeline, it looked perfect on paper:
documents → embeddings → vector search → LLM.

But the answers were wrong far too often.
The real problem? Naive retrieval strategies.

In this article, I break down practical techniques that dramatically improve RAG accuracy — including smarter chunking, query expansion, reranking, agentic retrieval, and more.

If you're building with LLMs, RAG, or AI search systems, this will save you a lot of debugging time.

Read the full article here:
https://t.co/dZSG8pd0zC

Curious:
What’s the biggest challenge you’ve faced while building RAG systems?

[Tarun Kumar Singh @krtarunsingh]

𝐈 𝐣𝐮𝐬𝐭 𝐫𝐚𝐧 𝐚𝐧 𝐀𝐈 𝐦𝐨𝐝𝐞𝐥 𝐨𝐧 𝐦𝐲 𝐨𝐥𝐝 𝐥𝐚𝐩𝐭𝐨𝐩.

No cloud.

No expensive GPU.

No AI server.

Just a simple tool called AirLLM that loads model layers one at a time.

It’s not ultra-fast, but it proves something powerful:

Large AI models can run locally on normal hardware.

I wrote a quick guide on how it works and how you can try it yourself.

𝐑𝐞𝐚𝐝 𝐡𝐞𝐫𝐞: https://t.co/a2M8uWwiPr

#AI #MachineLearning #LLM #Python #OpenSource

[Tarun Kumar Singh @krtarunsingh]

𝐈 𝐣𝐮𝐬𝐭 𝐫𝐚𝐧 𝐚𝐧 𝐀𝐈 𝐦𝐨𝐝𝐞𝐥 𝐨𝐧 𝐦𝐲 𝐨𝐥𝐝 𝐥𝐚𝐩𝐭𝐨𝐩.

No cloud.

No expensive GPU.

No AI server.

Just a simple tool called AirLLM that loads model layers one at a time.

It’s not ultra-fast, but it proves something powerful:

Large AI models can run locally on normal hardware.

I wrote a quick guide on how it works and how you can try it yourself.

𝐑𝐞𝐚𝐝 𝐡𝐞𝐫𝐞: https://t.co/a2M8uWwiPr

#AI #MachineLearning #LLM #Python #OpenSource

[Tarun Kumar Singh @krtarunsingh]

@elonmusk https://t.co/HeqUACn9x2

[Tarun Kumar Singh @krtarunsingh]

https://t.co/HeqUACn9x2

[Tarun Kumar Singh @krtarunsingh]

@elonmusk https://t.co/jSTonD09PS

[Tarun Kumar Singh @krtarunsingh]

@Heyyanyaa https://t.co/jSTonD09PS

[Tarun Kumar Singh @krtarunsingh]

https://t.co/jSTonD09PS

[Tarun Kumar Singh @krtarunsingh]

@elonmusk https://t.co/jSTonD09PS

[Tarun Kumar Singh @krtarunsingh]

First people chased apps.
Then they chased models.
Now they’re chasing agents.

[Tarun Kumar Singh @krtarunsingh]

@elonmusk 2026 attention span: 0.3 seconds
Grok: 'I got you, here's the whole video in 12 words' Humanity: crying in productivity @grok drop a 1-sentence summary of your life story pls