zsh · anton@buenos-aires:~/kovalev.dev — 120×40
live·day 16/90·
anton@buenos-aires:~$▍
Frontend engineer (10y) learning to build AI products in public.
RAG → Agents → Evals. Three projects in ninety days.
//journey / decrypting
DAY00/90
started 2026-04-14→target 2026-07-13·T-74 days to offer
//why this site exists
I spent a decade writing frontend for other people's products. I got good at it — and restless. The next ninety days I'm going deep on LLM apps: retrieval, agents, evaluation. Not reading about it. Building it, out loud, with receipts.
This page is the receipt. Three projects, one a month, each harder than the last. When the counter hits 90, I want an offer at an AI startup I respect.
No affiliate links. No course to sell. No "10x your output" thread. Just the work, shipped and timestamped.
idxpermsnameetastatusdescription
01drwxr-xr-xchat-with-docsday 15[shipped]RAG pipeline over PDFs. Embeddings, cosine retrieval, LLM reranking, query expansion, streaming, multi-turn.Next.jsClaude APIEmbeddingsTypeScript02drwxr-xr-xresearch-agentday 60[building]Agentic loop with Claude — tool selection, multi-turn execution, evals.Claude APITypeScriptEvals03drwxr-xr-xai-code-reviewday 90[queued]Reviews PRs the way a senior would — with taste, not rules. Runs on GitHub Actions.Claude APIGitHub ActionsTypeScriptEvals//learning / cat ~/syllabus.md
month 1 · apr
Claude API, prompting, RAG, embeddings, reranking, query expansion
month 2 · may
Agents, tool use, multi-turn evals, traces
month 3 · jun
Polish, portfolio, outreach, on-site interviews
background · 10y
React, Next.js, TypeScript, design systems, perf, DX