Hi, I'm Vadim Alakhverdov
Software developer writing about JavaScript, web development, and the tools that make us productive.
Recent Posts
Friday 12/06/2026
· 9 min readWhy Your "Working" AI Demo Will Break in Production: A Reality Check for PMs and Founders
The AI demo vs production gap, explained for PMs and founders: eight specific ways a working prototype breaks at scale, plus a pre-roadmap checklist.
Wednesday 10/06/2026
· 11 min readPricing AI Features: A Founder's Guide to Per-Seat vs Usage-Based Models
Per-seat vs usage-based pricing for AI features, with real gross-margin math for three products and a TypeScript calculator you can adapt to your own numbers.
Monday 08/06/2026
· 13 min readOpenAI Realtime API vs Gemini Live vs Pipecat: Which for Voice AI in TypeScript
Build the same voice agent in OpenAI Realtime, Gemini Live, and Pipecat-JS. A TypeScript comparison of latency, interruption handling, mid-stream function calling, and cost per minute.
Friday 05/06/2026
· 10 min readHybrid Search That Actually Works: BM25 + Embeddings + Reranking in TypeScript
Pure vector search misses exact-match queries like SKUs and error codes. Fix retrieval with hybrid search in TypeScript: BM25 + embeddings, fused with RRF, then reranked.
Wednesday 03/06/2026
· 11 min readPII Redaction Middleware: Strip Sensitive Data Before It Reaches the LLM
Build TypeScript middleware that redacts PII before it reaches the LLM, then re-hydrates the response - reversible redaction, audit logging, and a test suite.
Monday 01/06/2026
· 10 min readBuild an AI Eval Suite with Promptfoo: Catch Prompt Regressions Before Production
How to use Promptfoo to set up a TypeScript AI eval suite that catches prompt regressions in CI - deterministic asserts, LLM-as-judge, cost budgets, and a GitHub Actions workflow.
Friday 29/05/2026
· 11 min readWhy Your AI Feature Needs a Job Queue (And How to Add One with BullMQ)
Why synchronous LLM calls in Next.js API routes break under real load - and how to refactor to a BullMQ job queue with idempotency, priority lanes, SSE progress, and graceful shutdown.
Wednesday 27/05/2026
· 14 min readAdd AI Image Generation to Your Next.js App with Replicate, Fal, and Cloudflare R2
Add AI image generation to your Next.js app end-to-end - call Replicate or Fal with Flux, store outputs in Cloudflare R2, serve via signed URLs, handle webhooks, deduplicate, and budget cost per image.
Monday 25/05/2026
· 10 min readBuild Generative UI with Vercel AI SDK: Stream React Components from an LLM
Build generative UI with the Vercel AI SDK - stream real React components from an LLM tool call, render partial state during streaming, and gracefully fall back when a tool isn't registered.
Friday 22/05/2026
· 11 min readPinecone vs Turbopuffer vs pgvector: Which Vector Database for Production RAG in 2026
A real benchmark of Pinecone, Turbopuffer, and pgvector for production RAG in TypeScript. Latency, recall, and monthly cost at 1M, 10M, and 100M chunks.
Stay updated
Get notified when I publish new posts. No spam, unsubscribe anytime.