LLM Cost Optimization: Cutting Cost Per AI Interview
LLM cost optimization in a real product: per-stage model choice, prompt caching that survives, context trimming, and measuring quality before you cut.
All articles
Interview PrepFirst 5 Minutes of a Coding Interview: What to Do
What to do in the first 5 minutes of a coding interview: the clarifying questions worth asking, how to restate the problem, and how to agree an approach.
10 min
Interview PrepHow to Think Out Loud in an Interview Without Rambling
How to think out loud in an interview without rambling: a four-move structure for narrating your solution, what to say when stuck, and what loses marks.
10 min
EngineeringReal-Time Observability: What to Log for SSE and Voice
Real-time observability needs different signals than a CRUD API. What to log for an SSE stream or a live voice session, and which metrics predict bad UX.
9 min
EngineeringLLM Rate Limiting: Per-User Budgets and Cost Ceilings
LLM rate limiting has to protect a bill, not just a database. Per-user and per-IP budgets, per-session cost ceilings, and the one 429 you must never retry.
12 min
Interview PrepMock Interviews vs LeetCode: What Converts to Offers
Mock interviews vs LeetCode: what each one actually trains, which four criteria a scorecard weights, and how to tell which of the two is rejecting you.
9 min
Interview PrepHow to Practice Coding Interviews Alone: A Real System
How to practice coding interviews alone when you have no partner: six methods that build the out-loud skills interviewers score, and what each one cannot fix.
10 min
Interview PrepCoding Interview Evaluation Criteria: The Real Rubric
The coding interview evaluation criteria on a real scorecard, dimension by dimension, and what a weak and a strong answer actually look like on each one.
11 min
EngineeringRedis Date-Keyed Cache: Shipping a Problem of the Day
A Redis date-keyed cache looks trivial until midnight moves. How we key a daily coding challenge, set absolute expiry, seed safely and survive eviction.
11 min
EngineeringDeploy Frontend and Backend Separately: 6 Hard Edges
Deploy frontend and backend separately and six things break: cross-origin cookies, build-time URLs, proxy buffering, read timeouts, restarts, supervision.
12 min
AI NewsRun Muse Spark Open Locally: The Real Hardware Bill
Meta says you can run Muse Spark Open locally on a Mac. You need two of them, over 1TB of memory, and $21,000 — or just $1.25 per million input tokens.
7 min
AI NewsClaude Haiku 5.5 Pricing: The 100K Token Cliff
Claude Haiku 5.5 pricing starts at $0.10 per million input tokens, but cross a 100,000-token prompt and every single rate on the card jumps five times over.
7 min
EngineeringRazorpay Webhook Verification and Payment Idempotency
Razorpay webhook verification, idempotency keys, and reconciling a payment whose callback never arrived: the three server-side rules a checkout needs.
10 min
EngineeringRefresh Token Rotation Without Breaking Long Sessions
Refresh token rotation for a 60-minute session: how we choose access token lifetimes, keep an interview alive, and make logout mean something server-side.
10 min
EngineeringBrowser-Based Proctoring: Face and Gaze Signals in JS
Browser-based proctoring with MediaPipe: how we read face, gaze and head-turn signals on the client, batch them into events, and keep every signal advisory.
11 min
EngineeringRedux Toolkit Re-Renders: What We Keep Out of State
Redux Toolkit re-renders traced through a live AI interview session: what belongs in the session slice, what stays in a ref, and why the audio path skips React.
9 min
EngineeringProgrammatic SEO in Next.js: JSON-LD and Sitemaps
Programmatic SEO in Next.js from a real App Router codebase: a cross-linked JSON-LD @graph, sitemap.xml generated from data, and OG images built with sharp.
10 min
EngineeringNext.js Server vs Client Components: Where to Draw It
Next.js server vs client components decided in a real codebase: where we put the boundary, what generateStaticParams prerenders, and the bundle we measured.
10 min
EngineeringLLM as a Judge Rubric: Scoring Answers Reproducibly
A free-form score out of 10 is not a score. The LLM as a judge rubric design we use to make interview scoring reproducible, auditable and stable across runs.
11 min
Career & EdTech NewsFrontier Deployed Engineer: Inside Claude Frontier Academy
Anthropic will train 10,000 frontier deployed engineers by 2027. What the role actually involves, who gets nominated, and how to build the profile yourself.
7 min
AI NewsGemini 4 Argon vs GPT-6.1 Sol: Same Price, Different Bet
Gemini 4 Argon vs GPT-6.1 Sol: two frontier coding models at the same $2/$10 price. We compare DeepSWE v1.1 scores, cache rates and who can use them today.
7 min
AI NewsClaude Sonnet 5.5 vs Opus 5.5: Which to Use for Coding
Claude Sonnet 5.5 vs Opus 5.5: Sonnet is half the price and wins Terminal-Bench. Here is where Opus still wins, and why the real cost gap is 43%, not 50%.
8 min
EngineeringAI Interviewer System Prompt: Rules That Actually Stick
An AI interviewer system prompt alone won't stop a model lecturing or leaking the answer. Here are the config limits and tools that make the rules stick.
10 min
EngineeringVoice Agent Interruption Handling: Barge-In in Production
Voice agent interruption handling in production: how barge-in is detected, the three queues you cancel when it fires, and keeping the transcript honest.
10 min
EngineeringWebRTC vs WebSocket for Voice Agents: We Ship Both
WebRTC vs WebSocket for voice agents, from a product running both in production: the latency budget, audio-only negotiation, ICE and TURN, and barge-in.
9 min
EngineeringCodeMirror 6 in React: Language Switching and Bundle Size
Our CodeMirror 6 React editor rebuilt its whole extension tree on every keystroke. Here is the cause, the compartment fix, and what 11 languages cost.
10 min
EngineeringCode Template Wrappers: How Online Judges Run Your Code
A code template wrapper lets candidates write one function while the harness handles stdin parsing and output formatting across four languages, and what breaks.
9 min
AI NewsClaude Opus 5.5 Breaking Changes: 6 Errors, 2 Silent
Claude Opus 5.5 breaking changes go past the usual four: six request settings now return 400, and two more change how your app behaves without any error.
8 min
AI NewsClaude Opus 5.5 Pricing: What It Really Costs to Run
Claude Opus 5.5 pricing is $4 and $20 per million tokens. Here is how cache reads, effort defaults and batch mode change what an agent run really costs.
8 min
Career & EdTech NewsThe Tech and Management Job Market in 2026: Where the Opportunities Are
A data-backed look at tech and management hiring in 2026, with the latest figures from India and the U.S. and practical advice for job seekers.
9 min
EngineeringJudge0 Batch Submission: Streaming Per-Test-Case Results
Judge0 batch submission plus per-test-case callbacks beats N synchronous requests. How we submit once, stream partial results, and assemble them in order.
9 min
EngineeringRunning Untrusted Code Safely: A Sandbox Design Guide
Running untrusted code safely takes layers: real isolation, no network access, enforced resource limits, and a runner the public internet can never reach.
10 min
EngineeringServer-Sent Events with Redis: Streaming Test Results
How we stream code-execution results to the browser using Server-Sent Events with Redis behind them: keying run state, reconnecting a tab, closing cleanly.
10 min
EngineeringAWS VPC Design — Learning Roadmap
A 22-step roadmap for learning AWS VPC design from scratch — from CIDR and subnetting all the way to multi-AZ, production-grade network architecture.
16 min
Interview PrepAI Interview App: Complete Guide to Ace Your Tech Interviews
Learn how AI interview apps can help you prepare for technical interviews with real-time feedback and practice sessions.
9 min
Interview PrepBest Free AI Mock Interview Apps (2026 Guide)
A practical, no-nonsense look at the best free AI mock interview apps for practicing coding and behavioral interviews — what each is good at, and where the free tier ends.
7 min
Interview PrepHow to Prepare for Interviews Using AI Tools
Step-by-step guide on using AI interview tools to prepare for your next technical interview effectively.
7 min
EngineeringBuild a Browser Voice Agent: How Gemini Types Give Context to Your App
Learn how to use Gemini's tool calling and session control signals to add context and agentic features to your UI without MCP or LangChain.
5 min
EngineeringScaling Real-Time Apps: Best Practices for WebSocket Infrastructure
WebSockets can be costly for your infrastructure. Here are the core practices I used at Taqari.com to ensure a solid and secure communication layer.
7 min
EngineeringOpenAI Realtime API: Build Voice Agents with WebRTC (Beginner's Guide)
Learn how to use OpenAI's Realtime API with WebRTC to build voice agents directly in the browser. A complete beginner-friendly guide with code examples.
12 min
EngineeringUnderstanding Time Value and Date Comparison in JavaScript
Learn what time value means in JavaScript and how to compare dates correctly using time values.
7 min
Reading about interviews only gets you so far.
Practise out loud with an AI interviewer that asks real follow-up questions and tells you exactly what to fix. Free to start.
Start a free mock interview