Bappa 🌺🇺🇸🇮🇳 retweeted
hope all of you preparing for interviews, know this website
techinterviewhandbook.org
Bappa 🌺🇺🇸🇮🇳 retweeted
Learn LLM inference, build real stuff, and thank me later
All you need to get started
> kipp.ly/transformer-inferenc…
>huggingface.co/docs/transfor…
> docs.vllm.ai/
> github.com/sgl-project/sglan…
> arxiv.org/abs/2205.14135
Bappa 🌺🇺🇸🇮🇳 retweeted
Six undergrad schools (Berkeley, MIT, Harvard, Stanford, IIT, Columbia) produced about a quarter of US CEOs. The rest came from everywhere.
13% have a YC link. It's still the largest concentration of founder talent in one place; no other single place gives us this much signal. (7/8)
Bappa 🌺🇺🇸🇮🇳 retweeted
Don't waste 2 years learning to become an AI agentic engineer in 2026.
Andrew Ng, the godfather of AI, gave the complete playbook to become one from scratch.
1 hour course. Free:
• 00:00 - AI agent basics
• 12:12 - AI Agentic workflows & design patterns
• 53:27 - Practical tips for building AI agents
• 1:20:30 - self-improving AI agent loops
• 1:30:19 - multi-agent AI systems
I watched it last night.
Halfway through, I realized I could get into Anthropic in weeks, not years.
Bookmark now. Watch it. Then build your own AI agent
Bappa 🌺🇺🇸🇮🇳 retweeted
#WATCH | Breaking records. Challenging stereotypes. Inspiring change.
Indian skater Shrishti Dharmendra Sharma has broken her own Guinness World Record, completing a 50-metre limbo skating challenge beneath descending poles in just 6.86 seconds.
The medical student is using her record-breaking performances to challenge gender discrimination and inspire greater opportunities for girls.
#ShrishtiSharma #GuinnessWorldRecords #WomenEmpowerment
Bappa 🌺🇺🇸🇮🇳 retweeted
LLM engineer's handbook
(30 minutes a day, 10 weeks, 50 lessons)
a roadmap for llm inference serving where everything points at one service instead of scattering across demos. you get the mental model first, then serve a model, instrument it, load test it past 1000 concurrent requests, and tune it. you finish with a stack you configured yourself and a benchmark worth publishing.
here is what it covers:
→ the roofline model, and why decode waits on memory while prefill waits on compute
→ vLLM internals, PagedAttention and the scheduler, read from the code
→ Prometheus and Grafana for TTFT, inter-token latency, and queue depth
→ SGLang and RadixAttention prefix reuse, benchmarked against vLLM
→ load testing past 1000 concurrent requests
→ quantization across FP16, FP8, and INT4, on quality as well as speed
→ speculative decoding and KV eviction, including where the gains disappear
→ disaggregated prefill and decode, deployed on Kubernetes
→ a cost, latency, and quality router with per-request token budgeting
→ publishing a reproducible benchmark
the roadmap on GitHub: github.com/patchy631/time-to…
(don't forget to star 🌟)
i am also writing an article for each major topic. the first one is out, on how a GPU actually works.
the article is quoted below.
Bappa 🌺🇺🇸🇮🇳 retweeted
𝗔𝘀𝗶𝗮’𝘀 𝗳𝗶𝗿𝘀𝘁 𝗔𝗿𝗴𝗲𝗻𝘁𝗶𝗻𝗲 𝗙𝗼𝗼𝘁𝗯𝗮𝗹𝗹 𝗘𝘅𝗰𝗲𝗹𝗹𝗲𝗻𝗰𝗲 𝗖𝗲𝗻𝘁𝗿𝗲 𝗼𝗽𝗲𝗻𝘀 𝗶𝗻 𝗞𝗼𝗹𝗸𝗮𝘁𝗮.
Kolkata has become home to Asia’s first Argentine Football Centre of Excellence, launched on November 13 by Amuji Sports with AFA International, an authorised Argentine Football Association (AFA) programme. It will bring Argentina’s football methodology to India through schools and academies with local clubs. AFA International lists 10 academies and three football schools globally.
The programme includes intensive training and international tournaments from U-10 to U-20. Promising players may be shortlisted for training and exposure in Argentina through evaluation sessions. AFA International’s existing academy programme serves ages 4–18, with technical, tactical and goalkeeper training aimed at developing young players.
Bappa 🌺🇺🇸🇮🇳 retweeted
最近在研究 LangChain 和 LangGraph,发现很多人对这两个东西的理解其实有点混乱。
有人觉得 LangChain 是旧框架,LangGraph 是新框架;也有人认为 LangChain 适合简单 Agent,LangGraph 才能做复杂项目。
但其实,这两个东西根本不是简单的替代关系。
它们解决的是不同层级的问题。
先说 LangChain。你可以把它理解成一个高层 AI Agent 开发框架。
它主要帮你把大模型、Prompt、Tools、上下文以及 Agent 执行循环组织起来。
比如你想做一个招聘 Agent,让模型根据用户的问题,自动查询岗位 JD、读取相关信息、调用工具,最后生成分析结果。
如果直接使用模型 API,你可能需要自己处理模型请求、工具调用、结果回传和下一轮推理。
而 LangChain 已经封装了不少相关能力,可以让你更方便地把这些组件连接起来。
它更关注的是,怎么快速构建一个具备模型推理和工具使用能力的 Agent。
但问题来了。如果你的业务不是简单地调用工具、返回结果,而是需要多个明确的执行步骤呢?
比如招聘系统需要先解析简历,再匹配岗位,生成评分报告,然后根据业务规则进入不同处理流程,最后等待 HR 审批。
而且任务执行到一半突然中断了,你还希望它能够从之前的状态继续执行,而不是所有步骤重新跑一遍。
这时候就轮到 LangGraph 了。
LangGraph 更接近一个有状态的工作流编排框架。
它最核心的三个概念就是 State、Node、Edge。
State 负责保存执行状态,Node 负责执行具体任务,Edge 负责决定任务下一步流向哪里。
你可以明确规定哪些步骤必须执行,哪些地方允许模型自主决策,什么时候进入条件分支,什么时候暂停等待人工审批。
它还提供 Checkpoint、Interrupt 等机制,让开发者能够设计状态持久化、暂停与恢复流程。
但这里有个特别重要的区别。
LangGraph 能帮你恢复工作流状态,不代表它自动解决了业务幂等问题。
比如 Agent 调用一个付费 API,实际上已经执行成功,但程序还没保存结果就崩溃了。
如果恢复后直接重新调用,依然可能产生重复扣费。
所以真正的生产系统,还需要自己处理外部任务 ID、幂等键、业务状态和错误恢复。
说到这里,其实就很容易理解这两个框架的关系了。
LangChain 更关注模型和工具怎么组织起来,让 Agent 具备执行任务的能力。
LangGraph 更关注任务执行过程怎么被组织、控制和管理。
而且现在 LangChain 的 create_agent 本身就是建立在 LangGraph 之上的。
两者完全可以配合使用。
比如用 LangChain 创建一个能够调用招聘工具的 Agent,再把它放进 LangGraph 的某个节点,由 LangGraph 管理整个招聘业务流程。
当然,这也不代表所有 AI 项目都必须使用这两个框架。
如果你的需求只是调用一次模型,提取几个字段,返回一份 JSON,直接使用模型 SDK 可能更加简单。
如果业务流程完全固定,普通的后端代码和状态机也可能足够。
我觉得现在很多人学 AI 开发,最容易犯的错误,就是还没想清楚业务到底需要什么,就急着引入各种 Agent 框架。
最后框架装了一大堆,工作流画得特别复杂,真正出了问题反而不知道从哪里排查。
说到底,LangChain 和 LangGraph 真正值得学习的,不只是它们的 API 怎么调用。
而是它们背后的模型工具调用、任务状态、条件路由、执行恢复和工作流设计思想。
你得知道什么时候应该让模型自主选择工具,什么时候应该使用确定性流程,哪些状态必须保存,哪些操作需要人工审批,以及任务失败以后应该怎么安全恢复。
这些东西即使以后换了框架,依然是非常重要的工程能力。
所以我的理解是:
LangChain 让你更方便地构建 Agent。
LangGraph 让你更精细地控制 Agent 的执行流程。
但真正决定一个 Agent 能不能交付的,不是你选择了哪个框架,而是你有没有把业务流程、状态、权限和异常处理设计清楚。
学框架当然重要,但更重要的是搞明白框架到底帮你解决了什么问题。
不然最后 LangChain、LangGraph 都会用了,却连为什么需要一个 Checkpoint、为什么工具调用不能无脑重试都解释不清楚,那其实还是停留在工具使用层面。
这也是我觉得做 Agent、FDE、企业 AI 落地,最值得长期积累的工程能力。
Bappa 🌺🇺🇸🇮🇳 retweeted
Interesting comparison video of BrahMos and Tomahawk missiles.
In Op Sindoor, which missile caused catastrophic damage to Pakistan?
A. BrahMos
B. Tomahawk
Comment below 👇
Bappa 🌺🇺🇸🇮🇳 retweeted
India’s missile capabilities:
🇮🇳 One truck. One missile.
A range of 5,000+ km.
India’s missile capabilities span an impressive range, from short-range Prithvi missiles to the long-range Agni-V.
A look at India’s missile ladder:
🚀 Prithvi: 150–350 km
🚀 BrahMos: 290–800 km
🚀 Nirbhay: Approximately 1,000 km
🚀 Agni-I to Agni-V: 700–5,000+ km
From tactical strikes to strategic deterrence, India’s indigenous missile programme continues to strengthen its defence capabilities.
Probably the most unique and correct definition of smart.
Jensen Huang resigned as Nvidia's president, CEO, and a member of the board, effective immediately.
CFO Colette Kress is interim CEO, and the search for a permanent successor includes internal and external candidates.
earlier this year, on the A Bit Personal podcast, he was asked who was the smartest person he'd met.
"People who are able to see around corners are truly, truly smart, and their value is incredible."
He said that sits at the intersection of being technically astute and having human empathy, the ability to infer the unspoken, "the unknowables."
"To be able to pre-empt problems before they show up, just because you feel the vibe."
That vibe, he said, comes from a combination of data analysis, first principle, life experience, wisdom, and sensing other people.
"And that person might actually score horribly on the SAT."
15
Bappa 🌺🇺🇸🇮🇳 retweeted
When Indian markets were seeing even minor corrections, this clown was tweeting non-stop, trolling the PM, FM and Modi government, and questioning India's entire growth story.
Now that the markets have bounced back, attracted fresh investment and added nearly $660 billion in market capitalisation, suddenly he seems to have swallowed Fevicol!
And that 53-day narrative? Conveniently ignores the broader market recovery and the multiple global factors influencing equity markets.
Ab ye chomu GDP release ki date bhi nahi batata!
RT if you think these self-proclaimed market experts were never really concerned about investors or the markets. They were only interested in political trolling and engagement.
Elect BJP for at least next 5 elections. And then it can be seen where we are.
India's poverty reduction is the second biggest in world history, just behind China's. India is now growing at around 6.5%, which is about the same as China in 2014. Manufacturing is growing right along side services.
More people should pay attention.
noahpinion.blog/p/india-is-q…
59
Bappa 🌺🇺🇸🇮🇳 retweeted
Andrej Karpathy put his whole AI course on GitHub for free.
7 repos, in the order to learn them
→ nn-zero-to-hero, his video course that builds neural networks from scratch
github.com/karpathy/nn-zero-…
→ micrograd, the math that makes AI learn, in a tiny amount of code
github.com/karpathy/microgra…
→ makemore, a small model that learns to invent new names
github.com/karpathy/makemore
→ minbpe, how AI chops your words into tokens
github.com/karpathy/minbpe
→ nanoGPT, train your own GPT model
github.com/karpathy/nanoGPT
→ nanochat, "the best ChatGPT that $100 can buy"
github.com/karpathy/nanochat
→ autoresearch, AI agents that run AI experiments on one GPU by themselves
github.com/karpathy/autorese…
284,000 GitHub stars between them.
Everything you need to understand how ChatGPT works is free and open source.
Bappa 🌺🇺🇸🇮🇳 retweeted
RAG vs. Graph RAG vs. Agentic RAG, clearly explained!
Standard RAG embeds documents into vectors and retrieves the most similar chunks via similarity search. For direct factual lookups, this works well.
But it breaks down when a query needs to connect facts spread across multiple documents. Similarity search retrieves individual chunks, not the relationships between them.
Graph RAG adds a knowledge graph layer on top.
→ During indexing, an LLM extracts entities and relationships from the documents.
→ During retrieval, the system traverses these connections instead of relying on embedding similarity alone.
This is what enables multi-hop queries.
Say a vector DB stores three facts about internal services:
↳ "The checkout service uses payments API."
↳ "The payments API runs on cluster-3."
↳ "Cluster-3 is scheduled for maintenance on Friday."
Someone asks: "Will the checkout service be affected by Friday's maintenance?"
Vector search can likely retrieve facts 1 and 3 because the query mentions "checkout service" and "Friday maintenance."
But it will miss fact 2, which connects the payments API to cluster-3.
That middle fact sits too far from the query in embedding space. It mentions neither "checkout" nor "maintenance," so it never makes it into the retrieved context.
A knowledge graph connects these as linked entities, and graph traversal finds the full path in one query.
Agentic RAG takes a different approach entirely.
Instead of a fixed retrieval pipeline, an LLM agent decides at query time which tools to invoke, which sources to query, and in what order.
Check the visual below to understand the three architectures thoroughly.
One thing to note here is that these three aren't levels of sophistication that you need to graduate through.
Instead, they solve different query types.
↳ Single-hop factual lookups → standard RAG
↳ Multi-hop relationship queries → Graph RAG
↳ Dynamic multi-source tasks with tool use → Agentic RAG
Once the right architecture is in place, the next leverage point is efficiency.
Most RAG architectures rely heavily on vector search, and that layer can be made 32x more memory-efficient using binary quantization.
I covered the full implementation in the article below.
👉 Over to you: Which RAG architecture are you running in production?
Bappa 🌺🇺🇸🇮🇳 retweeted
Our HT Story: An ex-US Navy admiral who aided India during Galwan and worked on the GE F414 jet engine deal just won a US House seat
Could Rear Admiral (Retd.) Eileen Laubacher be a key new ally for India on Capitol Hill?
hindustantimes.com/world-new…
Bappa 🌺🇺🇸🇮🇳 retweeted
This is Taylor Perkins explaining Sanatan Dharma to western audiences
One thing I live about US education system is their communication skills they learn in school
Look how easy he makes for closed minded monotheistic people understand Hindu culture and thought
Bappa 🌺🇺🇸🇮🇳 retweeted
Wishing everyone a very Happy Diwali! May the festival of lights fill your lives with happiness, prosperity, peace and new beginnings. May the light of goodness always shine bright! 🪔✨
#HappyDiwali #Diwali2026 #ShubhDeepawali #TheStatesman
Bappa 🌺🇺🇸🇮🇳 retweeted
America’s full of these brown Uncle Toms trying to ingratiate themselves with White Anglo-Saxon Protestant America that doesn’t have the intellectual capacity to understand why India is the world’s most organically secular country. More certainly than @RoKhanna’s adopted nation.
He is member of the Pakistan House caucus. That says it all.
Unreliable person.
He should concentrate on the rising anti- Indian/ anti- Hindu hate in the US rather than street protests in India.
I have challenged his views before. @RoKhanna































