Portrait of Yuri Bondarenko

Yuri Bondarenko

Senior software engineer, DevOps and SRE

Based in Israel. I love designing and building products end to end. Several of my own products are live today.Based in Israel. I build AI agents and the tools they use: MCP servers, memory, monitoring and publishing APIs. At work I build AI agents for operations.Based in Israel. Today I'm a senior SRE: Kubernetes, Istio, Terraform and CI/CD for a platform with tens of thousands of requests per second.Based in Israel. I'm a senior SRE: Kubernetes (GKE, EKS, OpenShift), Istio, Terraform, SLOs and on-call for a platform with tens of thousands of requests per second.Based in Israel. I'm a developer first: I started in C# and Java, led a team building a medical device, and today I write Python and TypeScript every day.

Software engineering
15+ years
DevOps and SRE
8+ years
AI agents
since 2025
See my projectsAsk my AI

now shipping: AgentVivanight shift: shipping AgentViva

~/ai

Ask my AI assistant

It answers from my real work, checks a job fit, and can change this page for you.

ai@ybonda.com● online
claude haiku 5.5 · cloudflare ai gateway
Assistant:

Hi! I'm Yuri's AI assistant. Ask me anything about his work, or paste a job description and I'll check the fit.

Chats go to Claude through Cloudflare and are saved for 90 days. Don't share private info.

try one

~/projects

Things I've built

~/aigregator: how it's built

How it works

pre-built pagesasks and paysadvisor boxfor agentssign-in, reviewstools, users, reviewsranks the picksfinds by meaningtool profilesadd and enrichpublish syncsVisitorssearch, compare, reviewAI agentsMCP free, REST paidStatic siteNext.js on WorkersTool advisorneed in, ranked tools outAgent APIMCP and x402 RESTAccount APIFastAPI on Cloud RunRanking modelTypeSafe JevSearch indexVectorize and KVMongoDBthe tool catalogContent agentsAI agents on Cloud Run
  • People
  • App
  • Service
  • AI
  • Data
  • Visitors to Static site: pre-built pages
  • AI agents to Agent API: asks and pays
  • Static site to Tool advisor: advisor box
  • Agent API to Tool advisor: for agents
  • Static site to Account API: sign-in, reviews
  • Account API to MongoDB: tools, users, reviews
  • Tool advisor to Ranking model: ranks the picks
  • Tool advisor to Search index: finds by meaning
  • Agent API to Search index: tool profiles
  • Content agents to MongoDB: add and enrich
  • Content agents to Search index: publish syncs

Stack

Frontend
Next.jsTailwindshadcn/ui
Backend
Cloudflare WorkersHonoFastAPI on Cloud Runx402
Data
MongoDB AtlasCloudflare KVVectorize
AI
AI agents on Cloud RunTypeSafe JevWorkers AI
Open aigregator.com (opens in a new tab)

~/ttsensei: how it's built

How it works

pre-built pageschatpicks the setupchat logwhere to buyaccountsimagesedit proposalsreads and writesnews, reviewsdaily shop pricesClaude Haikuthrough AI GatewayPlayersbrowse, compare, reviewSetup advisorLangGraph on WorkersNext.js siteOpenNext on WorkersPartner shopsprices and buy linksCloudflare D1accounts, reviews, chatsCloudflare R2images, prices, catalogCatalog APIHono on Cloud RunContent agentsAI agents on Cloud RunMongoDBblades, rubbers, players
  • AI
  • People
  • Service
  • App
  • Outside
  • Data
  • Players to Next.js site: pre-built pages
  • Next.js site to Setup advisor: chat
  • Setup advisor to Claude Haiku: picks the setup
  • Setup advisor to Cloudflare D1: chat log
  • Next.js site to Partner shops: where to buy
  • Next.js site to Cloudflare D1: accounts
  • Next.js site to Cloudflare R2: images
  • Next.js site to Catalog API: edit proposals
  • Catalog API to MongoDB: reads and writes
  • Content agents to MongoDB: news, reviews
  • Content agents to Cloudflare R2: daily shop prices

Stack

Frontend
Next.jsTailwindshadcn/ui
Backend
Cloudflare WorkersOpenNextBetter AuthDrizzleHono on Cloud Run
Data
MongoDB AtlasCloudflare D1Cloudflare R2
AI
Claude HaikuLangGraphOpenRouterAI agents on Cloud Run
Open ttsensei.com (opens in a new tab)

~/geondex: how it's built

How it works

MCP toolsREST callsadds creditsruns the toolchecks creditssaves checksasks the questionGoogle datareads your pagesCoding agentsClaude Code, Cursor, CodexDevelopersREST API and playgroundEdge appAPI keys and creditsBillingplans and credit packsDatabasechecks and historyTool registry17 tools, REST and MCPKey-value cachecredits, rate limitsFive AI enginesPerplexity to GrokSearch datarank, volume, backlinksYour websitepage audit, AI crawlers
  • People
  • App
  • Outside
  • Data
  • Service
  • Coding agents to Edge app: MCP tools
  • Developers to Edge app: REST calls
  • Billing to Edge app: adds credits
  • Edge app to Tool registry: runs the tool
  • Edge app to Key-value cache: checks credits
  • Tool registry to Database: saves checks
  • Tool registry to Five AI engines: asks the question
  • Tool registry to Search data: Google data
  • Tool registry to Your website: reads your pages

Stack

Product
REST APIMCP serverAgent skills
Backend
TypeScriptEdge runtime
Data
SQL databaseKey-value cache
AI engines
PerplexityChatGPTClaudeGeminiGrok
Open geondex.com (opens in a new tab)

~/agentviva: how it's built

How it works

adds monitorsMCP toolspublishesone per targetprobes on a timeris the reply real?saves stateswhen state changesCustomersdashboard and REST APIAI agentsMCP server and pluginAgentViva WorkerHono: site, API, MCPStatus pagesbadges, Is X down?D1 databasemonitors, states, proofMonitor objectsone Durable Object eachJudge modelClef on Workers AIAI tool or agentpage, JSON, canary, MCPAlertswebhook, Telegram
  • People
  • App
  • Data
  • Service
  • AI
  • Outside
  • Customers to AgentViva Worker: adds monitors
  • AI agents to AgentViva Worker: MCP tools
  • AgentViva Worker to Status pages: publishes
  • AgentViva Worker to Monitor objects: one per target
  • Monitor objects to AI tool or agent: probes on a timer
  • Monitor objects to Judge model: is the reply real?
  • Monitor objects to D1 database: saves states
  • Monitor objects to Alerts: when state changes

Stack

Backend
TypeScriptHonoZod
Hosting
Cloudflare WorkersDurable ObjectsCron Triggers
Data
Cloudflare D1
AI and alerts
Workers AIWebhooksTelegram
Open agentviva.dev (opens in a new tab)

~/flashapi: how it's built

How it works

refreshes tokensone HTTP callsaves the poststarts publishingstores mediaresultscanary goes firstuploads videoMeta fetchesToken refreshdaily cronMy productsTTSensei, AigregatorD1 databaseposts, encrypted tokensFlashAPI Workermerge, policy, validatePublish workflowone step per accountR2 mediapublic media linksYouTubevideo onlyMeta platformsFB, Instagram, Threads
  • Service
  • People
  • Data
  • App
  • Outside
  • Token refresh to D1 database: refreshes tokens
  • My products to FlashAPI Worker: one HTTP call
  • FlashAPI Worker to D1 database: saves the post
  • FlashAPI Worker to Publish workflow: starts publishing
  • FlashAPI Worker to R2 media: stores media
  • Publish workflow to D1 database: results
  • Publish workflow to Meta platforms: canary goes first
  • Publish workflow to YouTube: uploads video
  • R2 media to Meta platforms: Meta fetches

Stack

Backend
TypeScriptHonoZodOpenAPI
Hosting
Cloudflare WorkersWorkflows
Data
Cloudflare D1R2KV
Platforms
FacebookInstagramThreadsYouTube
Open flashapi.dev (opens in a new tab)

~/memex: how it's built

How it works

search, editcommandsMCP toolslocal HTTPstdiofinds by meaningtext to 384 numbersbuilt from the DBMebrowser and terminalAI assistantsClaude Code and othersWeb consolelocalhost, live searchCLIsave, share, exportMCP serverone per assistantMemory storethe one write pathOne SQLite filememories and vectorsEmbedding modelruns on my CPUHTML pagesrendered live, sandboxed
  • People
  • App
  • Service
  • Data
  • AI
  • Me to Web console: search, edit
  • Me to CLI: commands
  • AI assistants to MCP server: MCP tools
  • Web console to Memory store: local HTTP
  • CLI to Memory store
  • MCP server to Memory store: stdio
  • Memory store to One SQLite file: finds by meaning
  • Memory store to Embedding model: text to 384 numbers
  • Memory store to HTML pages: built from the DB

Stack

Core
GoCobra CLI
Data
SQLitesqlite-vec
AI
bge-small-en-v1.5hugot
Ways in
Web consoleCLIMCP server

~/memo: how it's built

How it works

commandsMCP over stdiosaves and findstext to vectorsMarkdown mirroroptionalMein the terminalAI agentClaude Code, Cursor, Codexmemo CLIremember, search, exportMCP serverlocal, 9 toolsMemoryStoretwo-step dedupclaude -prewrites the MarkdownSQLite filememories and vectorsEmbedding modelbge-small, pure GoObsidian vaultone .md per memory
  • People
  • App
  • Service
  • AI
  • Data
  • Me to memo CLI: commands
  • AI agent to MCP server: MCP over stdio
  • memo CLI to MemoryStore
  • MCP server to MemoryStore
  • MemoryStore to SQLite file: saves and finds
  • MemoryStore to Embedding model: text to vectors
  • MemoryStore to Obsidian vault: Markdown mirror
  • MemoryStore to claude -p: optional
  • claude -p to Obsidian vault

Stack

Core
Gomcp-goCobra CLI
Data
SQLitesqlite-vec
AI
bge-small-en-v1.5hugot
Works with
Claude CodeCursorCodexObsidian
Open github.com/ybonda/memo (opens in a new tab)

~/graphmod: how it's built

How it works

asks for a diagramdraw as dotreply with dotis it safe?safe dot onlyshows itopen linksaves filesMein the terminalPrompt hookadds drawing rulesClaudewrites a dot blockSafety checkno file reads, no linksRender hookfinds the dot blockGraphvizgvpr style, dot layoutInteractive pagehover, drag, zoomPicture in replyPNG, or SVG in the appCache folderfiles by content hash
  • People
  • App
  • AI
  • Service
  • Outside
  • Data
  • Me to Claude: asks for a diagram
  • Prompt hook to Claude: draw as dot
  • Claude to Render hook: reply with dot
  • Render hook to Safety check: is it safe?
  • Render hook to Graphviz: safe dot only
  • Render hook to Picture in reply: shows it
  • Render hook to Interactive page: open link
  • Graphviz to Cache folder: saves files

Stack

Mod
TypeScriptClaude Code hooks
Drawing
Graphvizgvpr
Output
PNGSVGHTML page
Terminals
GhosttycmuxkittyWezTerm
Open github.com/ybonda/graphmod (opens in a new tab)

~/keyhush: how it's built

How it works

your prompttool outputscans the textscans the resultswaps each secretclean texttells youMetypes or pastesToolsBash, Read, MCPPrompt hookon every messageTool hookon every tool resultThree checksformats, names, entropyYour terminalsays what was hiddenMarkerssalted hash, per sessionClaudesees only markers
  • People
  • Outside
  • App
  • Service
  • AI
  • Me to Prompt hook: your prompt
  • Tools to Tool hook: tool output
  • Prompt hook to Three checks: scans the text
  • Tool hook to Three checks: scans the result
  • Three checks to Markers: swaps each secret
  • Markers to Claude: clean text
  • Three checks to Your terminal: tells you

Stack

Mod
TypeScriptClaude Code hooks
Detection
Key formatsSecret namesShannon entropy
Markers
Salted FNV-1a hash
Tested on
13,744 real filesNo false alarms
Open github.com/ybonda/keyhush (opens in a new tab)

~/resume.md

The short version

Full CV on LinkedIn
years in software
15+
years in DevOps / SRE
8+
clouds in production
AWS + GCP
  • Products, end to end

    I love architecting and building whole products: the idea, the design, the code, the infrastructure, the launch and the SEO.

  • AI agents and their tools

    I build AI agents and the tools they use: MCP servers, memory, monitoring and publishing APIs. At work I build AI agents for operations.

  • Production at scale

    Senior SRE today: Kubernetes (GKE, EKS, OpenShift), Istio, Terraform, CI/CD, SLOs and on-call for a platform with tens of thousands of requests per second.

  • A developer first

    I started as a software engineer in C# and Java, led a team building a medical device, and still write real code every day, mostly Python and TypeScript.

Skills

  • Kubernetes
  • GCP
  • AWS
  • Terraform
  • Istio
  • CI/CD
  • Observability & SLOs
  • Python
  • TypeScript
  • Cloudflare Workers
  • MCP
  • AI agents

How I work

  1. rule 01

    Ship small, ship often

    One product, one job. Live first, polish second.

  2. rule 02

    Run it at the edge

    Workers, D1, KV, R2. Cheap, fast, and boring to operate.

  3. rule 03

    Agents do the dull part

    I build the tools that let AI agents remember, post and check things for me.

Say hi.

Want to talk about a project, a job, or table tennis rubbers? I answer email.