Morning pour · 09:00 IST
New this week
One release a day that a developer should know about, picked from what shipped in the last few days and checked against its sources. Watch the reel, then read the breakdown.
October 2026
· 0:86
Universal Clipboard: how iPhone text pastes on a Mac
Copy on the iPhone, paste on the Mac. Here is what actually happens between the two, and when it works.
· 0:58
WhatsApp profile photo and name are not proof of identity
A familiar photo and a name like Boss are just data anyone can set. The phone number is the only identity WhatsApp has.
· 0:60
Streaming latency: why live video runs behind TV
The delay on a live stream is deliberate. With example numbers, chunks and a player buffer add up to about sixteen seconds, and low-latency mode cuts it to about four.
· 0:54
Cohort gating and waiting room: Flipkart BBD early access
Early access on a big sale is load control. A cohort list, a waiting room and batch-by-batch admission decide who gets in first.
· 0:63
Frequently bought together: the item-to-item method
Count what shoppers bought together, correct for popularity, rank the results and cache one list per item.
· 0:55
OTP scam: why you never tell it to a bank caller
The OTP is the final approval for a payment, so whoever asks you to read it out is the scam. Here is how the trick works.
· 0:60
Server-side ad insertion: how Hotstar shows different ads
Hotstar doesn't pause the match for an ad. It stitches the ad into your own playlist, and that's why your friend sees different ads.
· 0:59
IRCTC waitlist prediction: how a WL 23 chance is estimated
The waitlist number alone doesn't tell you much. Days left before travel change the odds, and a simple model built on past tickets shows why.
· 0:61
Google Maps traffic: how it spots jams before you do
The sensors are the phones of drivers ahead of you. A queue, some servers and a model turn their pings into green and red roads.
· 0:52
Blinkit 10-minute delivery: forecast, dark store, short ride
The ten minutes are won before you tap order, through a forecast, a stocked dark store and a short last mile.
· 0:27
AI Jugaad 5: checking a rejected health insurance claim
In this example, a rejection letter's own dates contradicted it, and AI chat found the gap and drafted the recheck request.
· 0:61
Cricket ball speed gun: built with Claude in 54 minutes
Day 2 of vibe coding: find the ball, measure the distance, divide by time. The first test read 17 for a 140 ball.
· 0:55
UPI PIN: you never enter it to receive money
A PIN screen means money is going out. Here is how the fake refund scam uses that gap, and the one rule that breaks it.
· 0:33
Seat locking: how ticket apps stop double bookings
A waiting room, a seat lock with a timer, then payment. That is how a typical ticketing app keeps two fans from getting the same seat.
· 0:39
Scam SMS spam filter design: rules, model and reports
Blocking the word "prize" fails the moment a scammer writes "pr1ze". Here is a layered filter that holds up better, in interview-style system design.
· 0:35
Instagram Reels recommendation system, explained simply
A typical Reels recommender shortlists candidates, scores them with several guesses, and keeps learning from watch logs.
· 0:38
Search autocomplete: the Flipkart system design answer
Flipkart's suggestions don't come from searching on every keystroke. A 150 ms wait, a cache and a sorted prefix index do the work.
· 0:39
Live score fan-out: how one six reaches 50M phones
One writer per match, a replicated cache and many push servers. That is the shape of a typical live score system, and it delivers in seconds.
· 0:37
Big Billion Days sale system design: cache, scale, queue
A typical sale-day design reads from cache, scales servers with load, and puts orders in a queue before they reach the database.
· 0:40
Live streaming system design: an IPL final to crores
Players buffer three chunks, so 6s chunks mean a long delay. Here is how a typical setup cuts it to a few seconds without flattening the origin.
· 0:39
UPI idempotency: how a double tap debits once
One payment, one transaction ID. Here is why a second tap or a retry on timeout doesn't take your money twice.
· 0:48
Live rider tracking on a map: a system design walkthrough
GPS pings, Redis pub-sub plus a last-position key, and WebSocket push. A common system design answer, broken down piece by piece.
· 0:45
Chatbot photo uploads: where your image goes and how safe it is
A photo you upload passes through a gateway, storage, a vision encoder and chat logs. Here is each step and what you can control.
· 0:31
Check if an SMS is a scam by pasting it into any AI
A quick habit for suspicious texts. Paste the message into any AI, ask one question, and read the red flags it lists.
· 0:34
AI Jugaad 2: ask an AI why your electricity bill jumped
Share a photo or the details of your bill, ask one question, and let an AI read the units and check the slab for you.
· 0:34
Bol Ke Bharo: speak Hindi, fill an English PDF form
A weekend build for English forms that mummy can't fill: speak in Hindi, let an LLM pull out each field, and let pdf-lib fill the PDF.
· 0:32
Credit card minimum due: ask AI what it really costs
Paying the minimum stops late fees but not interest. One question to an AI chat, with your statement attached, shows the difference.
· 0:31
AI fact-check for family group forwards, AI Jugaad 4
Copy the forward, paste it into an AI chat, and get each claim checked plus a kind reply you can send back to the group.
· 0:56
Semantic photo search with CLIP in the browser, built in 29 minutes
Day 1 of a vibe coding challenge: one spin, one hour, no rerolls. The result runs CLIP in the browser with no backend and no API key.
· 0:32
Learn AI for free: five videos to watch in order
You don't need a paid course. Watch 3Blue1Brown, freeCodeCamp, MIT 6.S191 and two Karpathy builds, in this order.
· 0:43
Tool calling: how an LLM calls get_weather
The model never runs get_weather itself. It asks, your code runs it, and the loop repeats until the model answers in plain text.
· 0:40
JetBrains Mellum2: open 12B coding model, 2.5B active
Mellum2 has 12B parameters but wakes only 2.5B per token, and you can serve it yourself with vLLM.
· 0:42
How ChatGPT remembers your chat: stateless models explained
The model remembers nothing between messages. The app around it stores the history and rebuilds the prompt on every turn.
· 0:40
IPL win predictor: how the system design fits together
Live ball events become chase numbers, and a model trained on past IPL chases turns them into a win chance after every ball.
· 0:42
Tatkal booking system design: waiting room, rate limit, seat lock
A typical system design answer for the 10 AM Tatkal rush, turning lakhs of simultaneous clicks into one orderly line.
· 0:38
EmbeddingGemma 2 for local multimodal retrieval
Google's open-weight multimodal embedding model targets private media search in local and edge apps.
· 0:34
Claude Haiku 5.5 should stay swappable
Anthropic's new small Claude is worth retesting on coding, triage, computer use, and background workflows.