Olivia
Online
Swtnk
@skc4232
Joined March 2013
614 Following
49 Followers
3.7K Posts
Salary wale log 1 lakh 2 lakh 5 lakh 10 lakh 20 lakh 30 lakh 40 lakh 50 lakh ka tax de rahe hain to get nothing in return
Kya mil raha hain itna tax dekar??
Government kab tak unko aise chusti rahegi ?
Kabhi na kabhi toh ghussa futega hi?
Max 20% tax for everyone.
Tax rates should be 5% 10% 15% 20%
No cess No Surcharge
The third round of talks with the CJP delegation was held today. The delegation submitted a written proposal covering cases related to the agitation along with its Five-Point Charter.
The government assured that no action will be taken against the organisers and protesters, and no new cases will be registered in connection with the agitation. It also expressed condolences to the families of the students who lost their lives and assured them of the highest possible compensation as per the rules.
The government has responded positively to the Five-Point Charter on examination and education reforms and will continue discussions with the delegation. Students and youth have been requested to end the agitation peacefully and allow the dialogue process to continue.
Thank you so much for your resignation, Pradhan ji.
It takes a lot of courage to make such a bold decision.
The students who were protesting against you do not have a personal problem with you. The protests were the result of multiple failures in the education system that they witnessed under your supervision, which ultimately led to this situation.
Stepping down does not make you small; in fact, it is an act of bravery.
I wish you the best for whatever you do next. 🙏
Your decision has restored my faith in democracy!
Jai Hind 🇮🇳
IT'S A VICTORY OF DEMOCRACY
direct democracy... straight from the streets.
It's a victory of peace, patience & persévérance.
Congratulations CJP, Gen Z of the nation and thank you all citizens for shedding fear and the fear of fear and rising up from every corner of the nation.
FROM ACCOUNTABILITY, NOW TO REFORMS
आज हमारे देश में पेट्रोल 82 रुपए प्रति लीटर मिलना चाहिए जबकि सरकार हमसे 102 रुपए प्रति लीटर ले रही है।
आज शाम 4 बजे प्रेस कांफ्रेंस
#Breaking
NEET परीक्षा पर पंजाब सरकार का बड़ा फैसला
सीएम @BhagwantMann ने X पर पोस्ट कर दी जानकारी
20, 21 और 22 जून को NEET छात्रों के लिए पंजाब रोडवेज बसों में सफर मुफ्त- भगवंत मान
NEET परीक्षा देने जा रहे छात्र सिर्फ एडमिट कार्ड दिखाकर बिना टिकट यात्रा कर सकेंगे
अरविंद केजरीवाल से बातचीत में छात्रों ने यात्रा खर्च में मदद की मांग उठाई थी।
गरीब और दूरदराज़ के छात्रों को राहत देने के लिए पंजाब सरकार ने किराया माफ़ करने का निर्णय लिया।
@AamAadmiParty @AAPPunjab
मैं राजनीति में ताकत के लिए नहीं आया। मैंने ज़िंदगी भर ईमानदारी कमाई है। जब कोई बेईमान कहता है, तो चोट लगती है। इसलिए तय किया था कि देश के सामने निर्दोष साबित होकर ही बोलूंगा। आज दिल का बड़ा बोझ उतर गया।
आख़िर में अधर्म और अन्याय हारता है और सच ही जीतता है।
सत्यमेव जयते
आज मेरे मन में यह भाव आया है कि अब एक ऐसा प्लेटफ़ॉर्म बनाया जाए, जिससे लोगों को इस अभियान से जोड़ा जा सके।
मंत्री रहते हुए भी मेरे सामाजिक दायित्व हैं, इसलिए मैं भी इसमें सक्रिय रूप से भागीदारी करूँगा।
#OnePlantADay
#5YearsOfShivSankalp
This book alone can change your ML interview game🙀
If you're serious about AI, ML, or landing top-tier roles... this book is DIFFERENT.
• Real-world deep learning interview problems
• Step-by-step solutions
• Covers ML math, probability, CNNs, Bayesian DL & more
• Designed for people who actually want to understand, not just memorize
This isn’t basic theory.
This is the kind of prep that makes you walk into interviews confident.
And I’m giving it FREE to first 4500
How to get it :
1️⃣ Follow me MUST (So i can dm you)
2️⃣ Like this post
3️⃣ Comment “DL”
4️⃣ RT to help others level up
If you’re preparing for ML interviews in 2026, this could literally change your trajectory.
No fluff. No gatekeeping.
Let’s build killers in AI. 🚀🔥
आज से मैं तय कर रहा हूं कि अपने स्वागत में माला और स्मृति चिह्न नहीं लूंगा।
आप एक पेड़ लगाकर आना और उसकी फोटो लाना, मेरा स्वागत हो जाएगा।
#OnePlantADay
#5YearsOfShivSankalp
From my experience, system design interviews don’t fall apart in the deep dives(say advanced stages).
In fact, they fall apart much earlier during API design.
As a senior(doesn't matter if you are a junior as well), you start to understand API design isn’t about naming endpoints perfectly. (well, its also a part of it tbh, lol).
I wrote down 20 frameworks in my notebook that I use for any interview(obv. system design ones), and they never failed me.
[1]. Treat API design as one step, not the interview
- spend ~5 minutes max. Move on to data flow and scaling.
[2]. Start from entities, then design APIs
- list nouns first. APIs are operations on entities.
[3]. Use REST as the default
- client → server? REST over HTTP unless there’s a strong reason not to.
[4]. Model URLs as plural resources, not actions
- paths are nouns. Methods express behavior.
[5]. Know HTTP methods and idempotency
- GET, POST, PUT, PATCH, DELETE and when retries are safe(idempotency).
[6]. Choose the right place for inputs
- path = identity.
- query = filters.
- body = create/update data.
[7]. Keep response schemas high-level
- reference entities. no need to enumerate fields unless asked.
[8]. Use status code ranges under time pressure
- 2xx success.
- 4xx client errors.
- 5xx server errors.
[9]. Reach for GraphQL only when flexibility matters
- different clients need different data shapes.
[10]. Call out N+1 problem with GraphQL(this always bothers me more than anything)
- and mention batching / data loaders.
[11]. Mention field-level authorization in GraphQL
- resolvers can enforce access per field.
[12]. Use RPC / gRPC for service-to-service calls
- lower overhead. Strong contracts. Faster internal traffic.
[13]. Explain why gRPC isn’t used for public APIs
- browsers and partners expect HTTP + JSON.
[14]. Acknowledge when there is no external API
- some designs are purely backend workflows.
[15]. Always paginate list endpoints
- anything unbounded must be paginated.
[16]. Use cursor-based pagination for dynamic data
- avoid skips and duplicates under heavy writes.
[17]. Be explicit about authentication and authorization
- public vs protected endpoints. Roles matter.
[18]. Put identity in headers, not request bodies
- never trust userId from the client payload.
[19]. Pick a session strategy and state it clearly
- JWT or server sessions — show where identity lives.
[20]. Keep explanations shallow first, deep only if asked
- signal clarity
-drill down on demand.
I always feel API design isn’t about perfection. It is
- moving fast without losing correctness.
- a clean structure
- a clear trade-offs
--------------------
Check out my Java+Spring boot+Microservices+Design Patterns+System design ebook curated for interviews from here https://t.co/iH4Ung369Y
![makakmayum_sid's tweet photo. From my experience, system design interviews don’t fall apart in the deep dives(say advanced stages).
In fact, they fall apart much earlier during API design.
As a senior(doesn't matter if you are a junior as well), you start to understand API design isn’t about naming endpoints perfectly. (well, its also a part of it tbh, lol).
I wrote down 20 frameworks in my notebook that I use for any interview(obv. system design ones), and they never failed me.
[1]. Treat API design as one step, not the interview
- spend ~5 minutes max. Move on to data flow and scaling.
[2]. Start from entities, then design APIs
- list nouns first. APIs are operations on entities.
[3]. Use REST as the default
- client → server? REST over HTTP unless there’s a strong reason not to.
[4]. Model URLs as plural resources, not actions
- paths are nouns. Methods express behavior.
[5]. Know HTTP methods and idempotency
- GET, POST, PUT, PATCH, DELETE and when retries are safe(idempotency).
[6]. Choose the right place for inputs
- path = identity.
- query = filters.
- body = create/update data.
[7]. Keep response schemas high-level
- reference entities. no need to enumerate fields unless asked.
[8]. Use status code ranges under time pressure
- 2xx success.
- 4xx client errors.
- 5xx server errors.
[9]. Reach for GraphQL only when flexibility matters
- different clients need different data shapes.
[10]. Call out N+1 problem with GraphQL(this always bothers me more than anything)
- and mention batching / data loaders.
[11]. Mention field-level authorization in GraphQL
- resolvers can enforce access per field.
[12]. Use RPC / gRPC for service-to-service calls
- lower overhead. Strong contracts. Faster internal traffic.
[13]. Explain why gRPC isn’t used for public APIs
- browsers and partners expect HTTP + JSON.
[14]. Acknowledge when there is no external API
- some designs are purely backend workflows.
[15]. Always paginate list endpoints
- anything unbounded must be paginated.
[16]. Use cursor-based pagination for dynamic data
- avoid skips and duplicates under heavy writes.
[17]. Be explicit about authentication and authorization
- public vs protected endpoints. Roles matter.
[18]. Put identity in headers, not request bodies
- never trust userId from the client payload.
[19]. Pick a session strategy and state it clearly
- JWT or server sessions — show where identity lives.
[20]. Keep explanations shallow first, deep only if asked
- signal clarity
-drill down on demand.
I always feel API design isn’t about perfection. It is
- moving fast without losing correctness.
- a clean structure
- a clear trade-offs
--------------------
Check out my Java+Spring boot+Microservices+Design Patterns+System design ebook curated for interviews from here https://t.co/iH4Ung369Y](https://pbs.twimg.com/media/G8dyE3jbgAASwtZ.jpg)
There are 2 career paths in AI right now:
The API Caller: Knows how to use an API. (Low leverage, first to be automated, $150k salary).
The Architect: Knows how to build the API. (High leverage, builds the tools, $500k+ salary).
Bootcamps train you to be an API Caller. This free 17-video Stanford course trains you to be an Architect.
It's CS336: Language Modeling from Scratch.
The syllabus is pure signal, no noise:
➡️ Data Collection & Curation (Lec 13-14)
➡️ Building Transformers & MoE (Lec 3-4)
➡️ Making it fast (Lec 5-8: GPUs, Kernels, Parallelism)
➡️ Making it work (Lec 10: Inference)
➡️ Making it smart (Lec 15-17: Alignment & RL)
Choose your path.
(I will put the playlist in the comments.)
♻️ Repost to save someone $$$ and a lot of confusion.
✔️ You can follow @techNmak, for more insights.
If you’re a "ML Engineer" training deep models and treating every datapoint as equally important, you’re missing how learning actually progresses inside the network.
The basic idea: models don’t learn from all data equally.
So, here's Concept 6: Sample Difficulty and Gradient Alignment
For each sample (xᵢ, yᵢ), the model updates through its individual gradient:
gᵢ = ∇θ ℓ(fθ(xᵢ), yᵢ)
But the optimizer doesn’t follow gᵢ. It follows the batch estimate:
g = (1/B) Σ gᵢ
This means a sample’s impact on learning depends on how well its gradient direction aligns with others in the batch or dataset.
Formally, the alignment between samples i and j can be written as:
Aᵢⱼ = (gᵢ · gⱼ) / (‖gᵢ‖ ‖gⱼ‖)
High Aᵢⱼ means the two gradients point in similar directions → their updates reinforce each other. These are the easy or consistent samples that the model learns early.
Low or negative Aᵢⱼ means the gradients disagree → updates partially cancel out.
These are the hard samples: noisy labels, edge cases, ambiguous inputs, or underrepresented patterns.
This creates a predictable training pattern:
• Early training: gradients are strongly aligned across many samples → high signal → rapid loss reduction.
• Late training: gradients diverge more → lower signal → learning becomes dominated by a small set of conflicting or difficult samples.
A sample’s effective influence on the update isn’t just its gradient magnitude. It’s much more. Here:
importance ∝ ‖gᵢ‖ × alignment(i, batch)
That is why two samples with equal loss can affect the model very differently.
This principle shows up everywhere:
• Curriculum learning works because it optimizes alignment over time.
• Hard example mining focuses on low-alignment samples with useful signal.
• Focal loss reduces the overweighting of highly aligned “easy” samples.
• Noisy or mislabeled samples cause unstable late-stage gradients because their alignment is close to zero or negative.
tldr: models don’t learn from all data equally.
They learn the aligned structure first, and only later adjust to conflicting structure. Gradient alignment( not dataset size) decides which samples actually shape the parameters.
Taj Mahal the incredible Architectural marvel 🔥
393 years ago A symbol of love built in Agra by the Mughal emperor Shah Jahan.
CJI बी.आर. गवई पर जूता फेंकने की कोशिश सिर्फ़ एक न्यायाधीश पर नहीं बल्कि भारत की आत्मा पर हमला है, इस देश की न्यायपालिका पर हमला है।
दलित बेटा मेहनत और ईमानदारी से देश की सर्वोच्च कुर्सी तक पहुँचे- ये इन लोगों को बर्दाश्त नहीं।
इनकी हिम्मत तो देखो। इनके समर्थक खुले आम CJI को सोशल मीडिया पर धमकियाँ दे रहे हैं। इस तरह की राजनीति और गुंडागर्दी ये देश कभी बर्दाश्त नहीं करेगा।
Many many happy returns of the day @SatyendarJain
The man responsible for bringing and implementing many out of the box ideas like Mohalla Clinics in Delhi . Fondly known as Idea Minister in our Govt. May God bless u with lots of good health, happiness and success.
गांधी जी का सपना था एक ऐसा देश जहां हर बच्चे को अच्छी शिक्षा मिले, हर जरूरतमंद को अच्छा इलाज मिले और जहां सभी धर्म और जाति के लोग भाईचारे से एक साथ रहें।
आज राष्ट्रपिता महात्मा गांधी जी की जयंती पर हम उन्हें श्रद्धा से याद कर रहे हैं और उनकी प्रेरणा से उनके सपनों का भारत बनाने के लिए संकल्पबद्ध हैं।
मयूर विहार फेज़-3 में हमारे कोंडली विधायक @KuldeepKumarAAP जी द्वारा आयोजित भव्य रामलीला मंचन में आज शामिल हुआ।
भगवान श्रीराम का आदर्श जीवन हमें सदैव सत्य, त्याग, सेवा और धर्म के मार्ग पर चलने की प्रेरणा देता है। भगवान श्री रामचंद्र जी के आशीर्वाद से आप सभी के जीवन में अपार खुशियाँ और तरक्की आए।
सियावर रामचंद्र की जय 🙏
Most Popular Users
Elon Musk 
@elonmusk
241.1M followers
Barack Obama
@barackobama
119.1M followers
Cristiano Ronaldo
@cristiano
112.4M followers
Donald J. Trump
@realdonaldtrump
111.8M followers
Narendra Modi
@narendramodi
107.1M followers
Rihanna
@rihanna
98.1M followers
NASA
@nasa
92.2M followers
Justin Bieber
@justinbieber
91.3M followers
KATY PERRY
@katyperry
88.7M followers
Taylor Swift
@taylorswift13
82.6M followers
Lady Gaga
@ladygaga
74.1M followers
Virat Kohli
@imvkohli
71.5M followers
Kim Kardashian
@kimkardashian
70.3M followers
YouTube
@youtube
68.7M followers
Bill Gates
@billgates
64.5M followers
Neymar Jr
@neymarjr
64.4M followers
The Ellen Show
@theellenshow
62.4M followers
CNN
@cnn
61.8M followers
Selena Gomez
@selenagomez
61.8M followers
X
@x
60.8M followers