Open to roles and contract work

Meelis MikiverFull-Stack AI Engineer & SEO Specialist

I ship full-stack products with agentic AI: consumer apps, admin platforms, LLM pipelines and 3D games. Then I rank them on search engines and LLMs so people find them.

View the work Contact

GitHub telemetry

Always shipping

14,398

contributions in the last year

contributions per day, on average

100

days, longest shipping streak

Jul 26 to Nov 2

396

contributions on the busiest day

Oct 18

2022

actively shipping since

products in production

Live from github.com/mlsmi, private contributions included

LessMore

Selected work

Eight products in production. Six built solo, end to end. AI Angels and its admin dashboard: engineering lead of a three-person team. Commit and PR counts are mine, from GitHub (November 2026).

Mars Rally Championship

marsrallychampionship.com

Browser rally racing on Mars: deterministic physics, online multiplayer and server-verified leaderboards.

deterministic physics, every run re-simulated server-side: 60 Hz; deterministic physics, every run re-simulated server-side
players per online race, with spectators and chat: 8; players per online race, with spectators and chat
replays rendered to video in the browser: 1080p; replays rendered to video in the browser

1,771 commits · 355 PRs

Next.js
Three.js
Colyseus
WebCodecs
Supabase
Solana

AI Angels

aiangels.io

Engineering lead, 3-person team

AI companion platform on the web and Telegram. I lead its engineering: architecture decisions, code reviews and a three-person team.

AI companions with streaming chat and voice, some with their own Telegram bot: 566; AI companions with streaming chat and voice, some with their own Telegram bot
pages indexed by Google: 132,000; pages indexed by Google
Google clicks a day: ~4,000; Google clicks a day

4,862 commits · 932 PRs

Next.js
Supabase
Prisma
Vercel AI SDK
xAI Grok
Deepgram
Telegram bots

Payments: 4 gateway integrations for cards and crypto, routed per plan, with signed webhooks and reconciliation crons
Auth orchestration: email codes, Google and Discord OAuth, guest sessions that merge into the account on signup, Telegram account linking
Email automation: Mailgun, SendGrid and SMTP failover with health checks, localized code emails, drip and capture sequences
Chat and voice: streaming LLM chat with model A/B arms and provider failover; voice via Deepgram, ElevenLabs and realtime calls
Group chats: an LLM director picks which of two companions speaks, with a scoring fallback and an energy budget for companion-to-companion banter
Creator platform: users build and publish companions, set credit prices, and earn 70% of fan subscriptions plus tips, paid out in stablecoins
Create wizard: seven steps from AI image and video generation to personality, names and version history
Credit economy: per-action pricing, daily streaks and mystery gifts, server-side locked media, and a no-account Day Pass with claim codes
Memory and realism: facts extracted every turn, session summaries and a nightly profile, proactive messages, multi-bubble replies with reactions
Safety: a real-time classifier, hard blocks, a permanent minor lock, crisis lines for 173 countries and an 18+ gate
Languages: the app runs in 12 languages with right-to-left layout, and companions reply in each user's own language
Growth: database-driven A/B tests, referral rewards and a 50% affiliate program, push notifications and an installable PWA
Free and premium experiences: guests, free users and subscribers get different message caps, media access, content ratings and voice limits, all resolved server-side from one entitlement check, with separate trial limits
Upgrade journey: in-character premium pitches at code-chosen moments with a guard against false claims, refusal cards that route to premium or credits, A/B-tested pricing layouts, promo codes, trials and a cancellation flow

AI Angels Admin Dashboard

internal

Engineering lead, 3-person team

The operations platform behind AI Angels: an AI image and video studio, an automated companion generator, SEO audits against Search Console, pricing, outreach, cost tracking and a live user map.

admin pages across 7 areas, on 477 API routes: 89; admin pages across 7 areas, on 477 API routes
cron pipelines for generation, SEO and analytics: 16; cron pipelines for generation, SEO and analytics
SEO video compilations and 11,700+ images: 40,000+; SEO video compilations and 11,700+ images

788 commits · 249 PRs

Next.js
Supabase
Seedream
Seedance
Kling
ffmpeg
deck.gl
Recharts
Playwright

Companion generator: casting, reference images, photos and video, on two cron lanes so slow video renders never block the quick steps
AI studio: image and video across Seedream, Grok, Seedance and Kling, shared character reference sets, a safety interlock on explicit references, and a pre-generated media library for in-chat requests
A/B experiments shared with the live site: hashed bucketing for guests and accounts, traffic splits, 18 tests including 13 pricing-page layout variants, significance testing on the create flow
Pricing and credits: plan editing with per-plan payment routing, pricing-page layouts measured against real payments, credit packs, promo codes and a credit-economy P&L
Conversion: upsell notifications with view and click tracking, targeted popup flows, in-site partner ads with AI-generated creatives, and testimonials
Content: landing pages over 5 templates with an AI-agent API (drafts until approved), an on-page chat widget, and a competitor review guide
Outreach: AI agents research and score partner sites and draft emails, an investor-finding agent with an MCP API, affiliate recruiting and application review
Engagement: segmented in-app notifications and drips, capped web push with quiet hours, and Telegram bot management with broadcasts
SEO audit: 59,000 Semrush keywords against weekly Search Console data, with AI page-fit and index checks
Analytics: acquisition, a chat-health funnel, call and conversation analysis, LLM cost per model, a live user map, a chat moderation queue and payment drift detection

aiwizard

aiwizard.ai

AI tools directory where an LLM pipeline researches, writes and fact-checks every listing, and tool owners verify themselves with an embedded badge.

tools reviewed and categorized: 10,000+; tools reviewed and categorized
categories in a two-level tree: 1,000+; categories in a two-level tree
pages prerendered on every build, no ISR: 12,000; pages prerendered on every build, no ISR

555 commits · 111 PRs

Next.js
Tailwind
Supabase
DeepSeek
Claude
Firecrawl
Vercel crons

Badge verification: owners embed a badge, an automatic check confirms it (rendering JavaScript-built sites when needed), and a daily cron re-checks every verified tool; two misses unverify
Verified tools earn a followed link, a blue check and verified-first placement on every listing; owners claim listings with a personal code carried on the badge
Submit-a-tool pipeline: a submission waits for its badge, then an AI review job writes the listing and screenshots the site and its tutorial steps for admin review
Catalog pipeline: scrape, DeepSeek enrichment with Claude fallback, and import, with an originality gate that rejects any copied 10-word span
Semrush-driven category expansion: LLM assignment and fill passes grow a two-level tree of 1,000+ categories
Supabase security: RLS policies, column grants, SECURITY DEFINER RPCs and a moderation trigger guarding curation fields

Multiplayer RPG in a rebuilt Bay Area. Fight from the Mission past the Golden Gate Bridge, Stanford's Hoover Tower and Sand Hill Road to a boss on Alcatraz. Your rarest NFT goat becomes your character.

skills levelling to 99 from XP: combat, woodcutting, fishing, cooking: 8; skills levelling to 99 from XP: combat, woodcutting, fishing, cooking
enemy types, from Startup Intern and Crypto Degen to VC Associate and the Board Member boss: 14; enemy types, from Startup Intern and Crypto Degen to VC Associate and the Board Member boss
regions with ~20 modelled landmarks, 64 players per shared world: 24; regions with ~20 modelled landmarks, 64 players per shared world

2,217 commits · 392 PRs

Next.js
Three.js
Colyseus
Solana
Metaplex
Drizzle
Neon
Redis

buybitcoinsmart

buybitcoinsmart.com

How to buy bitcoin in every country on earth, with a Claude-powered assistant and refuter-checked translations in ten languages.

programmatic pages in 10 languages: 50,000; programmatic pages in 10 languages
page types: country guides, payment methods per country, exchange-by-country pages, 659 comparisons, wallet hubs, US states, tax and licence guides: 22; page types: country guides, payment methods per country, exchange-by-country pages, 659 comparisons, wallet hubs, US states, tax and licence guides
Google clicks three months after relaunch, zero paid SEO: ~140/day; Google clicks three months after relaunch, zero paid SEO

2,676 commits · 146 PRs

Next.js
Tailwind
Supabase
Claude
i18n

AirdropStop

airdropstop.com

Crypto airdrop directory rebuilt from a lost database, where an AI pipeline sources every fact to a quote.

verified listings: 317; verified listings
databases: the repo is the CMS: 0; databases: the repo is the CMS

272 commits · 15 PRs

Next.js
Tailwind
Firecrawl
Claude

Ilusalong24

ilusalong24.ee

Estonia's beauty clinic directory, prerendered in four languages with hand-tuned grammar for each.

editor-checked salons across 10 service categories: 400+; editor-checked salons across 10 service categories
languages with hand-authored place-name grammar: 4; languages with hand-authored place-name grammar
routes prerendered per build: 9,600; routes prerendered per build

436 commits · 141 PRs

Next.js
Supabase
next-intl
Vercel

AI-native, by default

One developer with a fleet of AI agents now outships a team. I work that way daily, and I build the pipelines and automations that let products run themselves.

Claude Code
Cursor
Codex
Grok
DeepSeek
Vercel AI SDK

Agentic coding, directed end to end
Claude Code, Cursor, Codex and Grok are daily drivers. The 14,000+ contributions this year were shipped by directing AI agents through real codebases: specs, guardrails, review gates and test suites that keep velocity honest.
LLM pipelines in production
Content systems where models research, write and attack each other's work before anything ships: a judge, writer and refuter pipeline sourcing every fact to a quote, review enrichment across 10,000+ tools, and ten-language translation with independent refuters.
Automations that run the business
In-house bots and cron agents doing the operating: new records rendered to video on a GPU runner and posted to X, on-chain events cross-posted to Telegram, badge verification sweeps, index submissions and content refreshes, all unattended.

Stack

AI Engineering

Agentic coding with Claude Code, Cursor, Codex and Grok
LLM content pipelines with judge, writer and refuter passes
Claude, DeepSeek, xAI Grok and Mistral via Vercel AI SDK
Streaming chat with model A/B arms and provider failover
AI image and video generation: Seedream, Seedance, Kling, Grok Imagine
Voice AI: Deepgram, ElevenLabs, realtime calls
Agent APIs and MCP servers for AI-written content

SEO & Growth

Programmatic SEO at 130k+ indexed pages
International SEO: 96 locales, reciprocal hreflang
Sharded sitemap indexes, video and image sitemaps
JSON-LD structured data across 15+ schema types
IndexNow, llms.txt and answer-engine optimization
Search Console and Semrush keyword pipelines
Internal-linking engines and thin-content gates

Frontend

TypeScript
React 19
Next.js 15/16 App Router
Tailwind CSS
next-intl
Radix and shadcn/ui
Recharts and deck.gl

Backend & Data

PostgreSQL
Supabase (Auth, RLS, RPCs)
Prisma
Drizzle and Neon
Upstash Redis
Vercel crons and background drains

Payments, Auth & Email

Multi-gateway card payments, routed per plan
Crypto payments with reconciliation crons
Telegram Stars and TON
Auth: email codes, Google and Discord OAuth, guest-to-account merge
Email: Mailgun, SendGrid and SMTP failover, drip sequences
PostHog analytics and OneSignal push
Firecrawl data harvesting

Realtime, 3D & Media

Three.js
Colyseus authoritative servers
Deterministic physics with server re-simulation
WebCodecs video rendering
ffmpeg video pipelines

Web3 & Telegram

Telegram Mini Apps
Telegram bots (Bot API)
Solana and Sign-In With Solana
Metaplex NFTs
On-chain event automation

Platform & Ops

Vercel at scale (ISR and build economics)
GitHub Actions incl. a self-hosted GPU runner
Playwright and Vitest
Railway WebSocket relays
Admin dashboards and internal tools

Contact

Hiring for a team, or need a product built and run end to end? Tell me what you are working on.

Based in: Estonia (EET)
Working: Remote, EU hours
Open to: Full-time roles and project contracts

github.com/mlsmi LinkedIn