Building a Real-Time Ad Platform Under a Millisecond Budget
Design a bidder that respects auction deadlines, handles stale features and duplicate events, bounds spending across regions, and degrades without bypassing permission.
TechScribr
Machine learning, made clear
Build intuition before complexity.
Practical, deeply explained guides to machine learning, statistics, deep learning, and the systems that put them to work.
Explore by topic
Curated sequences that build one idea at a time.
16 lessons Follow an ad through the system
Explore all learning paths AdTech from first principles
Trace the auction, infrastructure, machine learning, privacy constraints, and bidding decisions behind an impression. Start with The invisible auction 12 lessons From measurement to decisionsExperimentation and causal inference
Build an experiment correctly, choose an analysis, and understand what changes when randomization is unavailable. Start with P-values and confidence intervals 10 lessons From generation to efficient modelsPractical LLM systems
Learn how decoding, memory, retrieval, quantization, and parameter-efficient fine-tuning work under the hood. Start with LLM sampling techniques 11 lessons Start with the foundationsProbability distributions
Learn to recognize the random process behind a problem and choose a distribution that matches it. Start with A field guide to probability distributionsLatest writing
Clear explanations, practical insightsPractical Privacy on the Web: Reduce Exposure Without Giving Up the Internet
Choose privacy controls by what they actually change: browser requests, cross-site links, mobile permissions, and the use of data already collected.
The Data Exhaust of a Single Page Load: What Leaves the Browser, and What It Reveals
Follow a publisher page through requests, storage, identifier syncing, and inference, then learn how to audit the evidence without overstating what a browser trace proves.
Trusting the Programmatic Supply Chain: Follow the Seller, Then Question the Inventory
Trace ads.txt, sellers.json, and OpenRTB schain through a worked supply-chain investigation, then evaluate spoofing, inventory quality, and supply-path economics.
Campaign Lifecycle Inside a DSP: From Advertiser Brief to Bidding Decisions
Turn a ₹1,00,000 campaign into line items, targeting, creative approvals, pacing, frequency controls, and a measured optimization loop inside a DSP.
Anatomy of an OpenRTB Auction: The Request, the Bid, and What Happens Next
Read a worked OpenRTB request and response, trace the auction deadline and IDs, and separate a winning bid from creative delivery and billing.
One Impression, Two Ledgers: From Campaign Budget to Publisher Payment
Follow one ad opportunity through campaign rules, data sharing, delivery, billing, and settlement, with a worked ₹1,000 spend example.
Make the Agents Argue: Adversarial Review for Code That Cannot Fail Quietly
How a solo developer ships an AI-built options-market data and research system: two AI architects working in isolation, priced adversarial findings, human gates, and the bugs the process actually caught.
Prompt Caching: Paying Once for the Prefix that Every Request Repeats
How prompt caching reuses identical LLM input prefixes, why causal attention makes it possible, when the economics work out, and how to diagnose cache misses.
From Conversion Pixels to Incrementality: What Did the Ad Actually Cause?
A practical guide to attribution rules, conversion deduplication, randomized holdouts, and the gap between credited revenue and causal lift.