☕ Banner 219/367
Turning raw data into dashboards, pipelines into insights, and complexity into clarity
Three independent Cloudflare + Claude builds — the portfolio's own ask --live widget, a focused RAG practice lab grounded in 128 official USCIS civics questions + real SQL sessions, and a self-hosted mem0-style long-term memory API.
💡 I build end-to-end data pipelines and MLOps systems — from raw ingestion to production-ready dashboards — with a focus on Docker, CI/CD automation, DuckDB, and data storytelling that drives real business decisions.
🛠️ Core Stack
SQL • Python • PostgreSQL • DuckDB • Docker • GitHub Actions • MLflow • Tableau • Power BI• Excel
📊 Focus Data Engineering • MLOps Automation • Business Intelligence • ETL/ELT Pipelines • Agentic AI Engineering
⚡ How it works (architecture deep-dive 🔬 for engineers)
This profile is a self-updating MLOps demo — a living portfolio showcasing production-grade automation.
- 🤖 Banner rotation: 367 GIFs · natural sorting · cache-busted CDN URLs
- 🧩 Dynamic insights: Context-aware NLG (time/season/DOW algorithms)
- ⏱️ Next Update badge: Shields.io endpoint · HLS gradient · sub-minute precision
- 📡 Observability: JSONL telemetry · heartbeat pings · state persistence
- ⚙️ Zero-touch ops: 5,700+ scheduled runs · 18,200+ total CI events · 377 mutations · idempotent commits
| File | Version | Description |
|---|---|---|
| update_readme.py |  |
Banner engine + NLG + JSONL pipeline |
| build_next_badge.py |  |
HLS gradient renderer + countdown |
| build_activity_graph.py | |
GitHub GraphQL → self-hosted 30-day trend chart SVG |
| Workflow | Schedule | Runs | Status |
|---|---|---|---|
| Auto Update README | Daily 12:15 UTC | 3,063 |  |
| Next Update Badge | Every 20min | 8,660 |  |
| CI/CD Pipeline | On push/PR | 3,518 |  |
| Smoke Tests | Daily | 357 |  |
| Cache GitHub Trophies | Every 6h | 396 |  |
| Generate Snake | Daily 00:30 UTC | 346 |  |
| Extra Badges | Daily 10:30 UTC | 101 |  |
| Activity Graph | 05:15 & 17:15 UTC | new |  |
.
├─ update_log.jsonl # CI run timeline (1 JSON per run: ts_utc, run_id, run_number, sha, banner_*, insight_*)
├─ update_log.txt # Grep-friendly mirror of update_log.jsonl (ts UTC, run=…, sha=…; rolling tail)
├─ badges/
│ ├─ next_update.json # Live Shields.io badge state (label, message like '~14h 35m', color bucket)
│ ├─ next_update_log.jsonl # Badge countdown snapshots (ts, next_utc, minutes_left, message, color, jitter params)
│ ├─ next_update_log.txt # Human-readable badge ETA tail ([ts] color=… msg='…' next_utc=… mins_left=…)
│ ├─ github_followers.json # Endpoint payload for the Followers badge (schemaVersion/label/message/color)
│ ├─ github_stars.json # Endpoint payload for the Stars badge
│ ├─ total_updates.json # Endpoint payload for the Updates badge
│ ├─ trophies.svg # Cached GitHub Trophies SVG (via Cache GitHub Trophies workflow)
│ ├─ snake_variant.json # Active snake color variant (label/color, updated by snake.yml)
│ ├─ github_contributions.json # Total contributions this year (GraphQL, via badges_extra.yml)
│ ├─ github_commits.json # Commit count this year (GraphQL, via badges_extra.yml)
│ ├─ github_issues.json # Issues opened this year (GraphQL, via badges_extra.yml)
│ └─ activity_graph.svg # Self-hosted 30-day trend chart (GraphQL, via activity_graph.yml)
└─ .ci/
├─ heartbeat.log # GitHub Actions heartbeat ledger (Updated on / Triggered by / Commit SHA / Run ID / Run number)
└─ update_count.txt # Monotonic mutation counter (powers the «N mutations shipped» tagline)
📋 Browse logs: 📊 update_log.jsonl · 📝 update_log.txt · 💓 heartbeat.log · 🔢 update_count.txt ⏱️ next_update.json · 📡 next_update_log.jsonl · 📋 next_update_log.txt 👥 github_followers.json · ⭐ github_stars.json · 📈 total_updates.json · 🏆 trophies.svg · 🐍 snake_variant.json · 📊 github_contributions.json · 🔨 github_commits.json · 🐛 github_issues.json · 📅 activity_graph.svg
Focus
- 📊 Data Analytics & Business Intelligence
- 🧠 Advanced SQL, Data Modeling & Analytical Thinking
- ⚙️ Analytics Engineering · ETL/ELT workflows · Pipeline automation
- ☁️ Cloud Analytics — Azure Databricks, Data Factory, Synapse Analytics
- 🐍 Python & R for data science workflows
- 🤖 Agentic AI Engineering — Claude Code workflows for architecture, debugging & production incident response
🧭 2.5+ years delivering production data pipelines, live analytics dashboards, and automated MLOps workflows — from raw ingestion to deployed applications
- 🎓 SuperDataScience — Data Analytics, ML & Automation
- 📘 Udemy — SQL, Tableau, Power BI & Data Projects
- ☁️ CloudWolf — AWS & Azure fundamentals for data workflows
- Build dashboards that answer real business questions (Tableau, Power BI)
- Write advanced SQL — CTEs, window functions, optimization, not just
SELECT * - Design and automate ETL/ELT pipelines end-to-end (Python, PostgreSQL, DuckDB)
- Model data for analytics — star schema, dimensional modeling, data contracts
- Work with cloud analytics stacks (Azure Databricks, Data Factory, Synapse)
- Turn raw data into decisions — fast, reproducible, and production-grade
- Debug and ship production fixes with AI-agent tooling — verified against real logs and screenshots, not guesses
| Project | Highlights | Demo |
|---|---|---|
| 🧠 Agent Memory | Cloudflare Vectorize + Workers AI + D1 + Claude Haiku 4.5 · mem0/Supermemory-style memory API · atomic fact extraction · similarity × importance × recency re-ranking | 🌐 Live |
| 🦎 Civics × SQL RAG Lab | Cloudflare Vectorize + Workers AI (bge-m3) + Claude Haiku 4.5 · 128 official USCIS civics Q&A + real SQL practice sessions · single index, metadata-filtered by mode | 🌐 Live |
| 🚚 Route Optimization VRP | CVRPTW · Google OR-Tools · 100% vs 11% on-time · 18.7% distance saved · $82/day saved |  |
| 📦 CV Logistics MLOps | ResNet18 transfer learning · MLflow + W&B · val_mae 0.755 · weekly automated retraining | — |
| ₿ Crypto On-Chain Dashboard | Top 20 coins · Fear & Greed · BTC dominance · DuckDB · daily pipeline | |
| 📉 Tech Layoffs Tracker | 2,412 events · 747K people · 49 countries · DuckDB · weekly refresh | |
| 📊 SO Survey Analytics | 65K devs · 20 SQL queries · Remote +51% · DuckDB · 23 CI tests | |
| 🌍 Global Weather Pipeline | 20 cities · 6 continents · Best City Score · 7d Forecast · Quality Layer | |
| 🔌 MCP Data Quality Agent | 19 MCP tools · 5 databases · Claude AI · natural language analytics | — |
| 🎓 Data Interview Coach | 20 questions · SQL + Behavioral + Project · Claude API · streaming feedback · SQLite | |
| 📈 Job Market Pulse | 10 stacks · 10 US cities + remote · 110 API calls/day · DuckDB · daily pipeline | 🤗 Live |
| 🛒 Olist Analytics | dbt · 54 tests · $13.2M · 96K orders | 🤗 Live |
| 🚗 Uber Driver Analytics | 3,448 trips · $70K gross · 98.9% rating | 🤗 Live |
| ❄️ Snowflake A-Z | 11 hands-on practices · Snowpark · Dynamic Tables · RBAC · native Streamlit · SQLFluff CI | — |
| 🏢 HR BI Analytics | 30 employees · 5 depts · Sales $102K avg · Tableau | — |
| 📊 Business SQL Analytics | 2,314 cust · 5K transactions · $2.58M · 59.8% returning | — |
| 🦆 NYC 311 DuckDB | 22,504 records · Bronx 41.5% · DuckDB · MotherDuck | — |
| 🔄 ETL Pipeline | Faker → PostgreSQL · SQLFluff CI · Docker | — |
| 🌍 Remote Job Tracker | 100 listings · 5% remote · Munich 36% · API→Tableau | — |
| 🤖 MLOps Project | R²=0.8326 · RMSE $46K · MLflow + W&B · 729 GridSearch | — |
| 🧠 FastAPI + Ollama Playground | Local LLM inference · phi3 · llama3 · deepseek-r1 · streaming API · Docker Compose | — |
⚡ AI-Powered Engineering Workflow
| Assistant | Role | Usage |
|---|---|---|
| 🧠 Claude Sonnet 5 | Primary AI Partner — architecture · code · analytics · docs · review | Primary |
| 🖥️ Claude Code (CLI) | Agentic Runtime — full-auto sessions · persistent context · tool-verified changes | Primary |
| 🕵️ OpenAI Codex CLI | Second Opinion — blind code review · isolated one-off tasks · breaking stuck-debug loops | On-demand |
💡 How Claude fits into my workflow
Claude Sonnet 5 is my primary AI engineering partner across all stages of the data & MLOps lifecycle:
- 🏗️ Architecture → pipeline design, schema decisions, project structure
- 🐍 Code → Python scripts, SQL queries, Docker configs, GitHub Actions workflows
- 📊 Analytics → data modeling, query optimization, business logic translation
- 📝 Documentation → READMEs, project descriptions, technical write-ups
- 🔍 Review → debugging, code quality, edge case analysis
- 🚨 Production Ops → live incident triage from real logs/screenshots — root-cause diagnosis, verified fix, deployed and re-checked in-session
Precision-first · Context-engineered · Production-grade output.
🕵️ How Codex CLI fits into my workflow
OpenAI Codex CLI is my second opinion — deliberately kept blind to Claude's project context, so it reviews cold instead of just echoing conclusions I've already reached:
- 🔍 Blind code review → reviews a diff/module with zero access to my CLAUDE.md conventions or fix history — catches what a context-loaded reviewer misses
- 🌐 Isolated translation → one-off conversions (e.g. SQL dialect ports) where the output is easy to verify standalone
- 🧯 Breaking stuck-debug loops → brought in fresh, with no memory of failed attempts, when the primary agent is stuck re-fixing the same bug
- 🌳 Isolated via
git worktree→ runs on its own branch in its own directory, never alongside the primary agent in the same working tree, so neither can silently overwrite the other's edits
First real find (2026-09-20): a blind Codex review of
route-optimization-vrp's OR-Tools segfault-isolation code surfaced two real bugs — a timeout that didn't actually kill a hung worker, and no cap on concurrent solves — that 5 prior fix commits had missed. Verified against the actual code, fixed, and shipped in PR #1, confirmed clean in production logs post-deploy.
🛰️ AI Systems in Production
🤖 RAG Ask-Widget — Live on This Profile
An interactive Q&A widget embedded on this page, answering visitor questions about my projects and background in real time — a shipped AI product I designed and deployed end-to-end, not just a tool I use for work.
| Component | Implementation |
|---|---|
| 🔎 Retrieval | Cloudflare Vectorize — 1024-dim index, cosine similarity, multilingual bge-m3 embeddings |
| 🧠 Generation | Claude (Anthropic API) — grounded, context-injected answers |
| 📥 Ingestion | Cloudflare Queues — async corpus embedding/upsert, dead-letter queue for failed jobs |
| 🛡️ Bot protection | Cloudflare Turnstile — verified before any rate-limit or generation cost is incurred |
| ⏱️ Rate limiting | Per-IP + global daily caps via Workers KV — cost-bounded by design |
| 📊 Analytics | Cloudflare D1 — anonymized topic/language classification per question |
| ⚡ Runtime | Fully serverless, edge-deployed on Cloudflare Workers — zero always-on infrastructure |
🌐 Live demo → · 💻 Source →
🔒 Privacy by design: analytics store only a keyword-classified topic bucket and detected language — never the raw question text.
🧠 agent-memory — Self-Hosted Memory Layer for AI Agents
A mem0/Supermemory-style long-term memory API — extracts atomic facts from raw text, embeds and stores them, then retrieves by relevance, importance and recency, not similarity alone. A standalone shipped AI product (separate repo), built on the same Cloudflare pattern as the ask-widget above, not a wrapper around a third-party memory service.
| Component | Implementation |
|---|---|
| 🧠 Extraction | Claude Haiku 4.5 — pulls typed, atomic facts (preference/fact/event/correction) with an importance score out of raw text |
| 🔎 Retrieval | Cloudflare Vectorize — 1024-dim index, cosine similarity, bge-m3 embeddings |
| 🗂️ Storage | Cloudflare D1 — structured metadata (type, importance, timestamps) |
| ⚖️ Re-ranking | similarity × importance-weight × recency-decay — an important fact from weeks ago still outranks a trivial fresh one |
| ⚡ Runtime | Fully serverless, edge-deployed on Cloudflare Workers |
🌐 Live demo → · 💻 Source →
🎨 Creative & Content Generation
Portfolio banner visuals for all 20 project cards were generated with Gemini (nano banana), refined through iterative prompting — concept → test batch → visual QA (composition, palette, icon accuracy) → full rollout. Design/marketing tooling only — engineering work stays on the Claude stack above.
🖼️ See the generation process (3 examples)
MLOps Docker — real Docker whale icon instead of a generic shield, two-panel layout
Snowflake — single-panel + floating badge layout, brand-accurate snowflake mark
Route Optimization — three-panel cascading layout, first fully-validated complex composition
🤖 Automation Logs
🪄 Run Meta (click to expand)
- 📆 Updated (UTC): 2026-09-27 17:22 UTC
- 🤖 Run: #5882 — open run
- 🧬 Commit: 91dd433 — open commit
- ♻️ Updates (total): 494
- 🌀 Workflow: Auto Update README · Job: update-readme
- ✨ Event: schedule · 🧑💻 Actor: evgeniimatveev
- 🕒 Schedule: 24h_5m
- 🌈 Banner: 219/367
🗂️Recent updates (last 5)
| Time (UTC) | Run | SHA | Banner | Event/Actor | Insight |
|---|---|---|---|---|---|
| 2026-09-27 17:22:08 | 5882 | 91dd433 |
219/367 (219.gif) | schedule/evgeniimatveev | 📡 TEST • OBSERVE • DEPLOY • RUN #5882 — Tune models, store wisdom 📦 | Prep for an MLOps-filled week! ⏳ Monitor, alert, respond — cal… |
| 2026-09-26 16:50:52 | 5881 | e61f5ea |
218/367 (218.gif) | schedule/evgeniimatveev | 📡 OBSERVE • ALERT • IMPROVE • RUN #5881 — Backtest Decisions, Bank Learnings 🏦 | Weekend Automation Vibes! 🎉 Review Metrics, Cut Toi… |
| 2026-09-25 17:42:07 | 5880 | 1c1189a |
217/367 (217.gif) | schedule/evgeniimatveev | 📡 MLOPS DAILY • RUN #5880 — Prune Legacy, Keep Essentials ✂️ | Wrap It Up Like A Pro! ⚡ Optimize, Deploy, Repeat! 🔄 🔄 |
| 2026-09-24 17:40:15 | 5879 | d64a042 |
216/367 (216.gif) | schedule/evgeniimatveev | 📡 ETL → FEATURES → IMPACT • RUN #5879 — Migrate legacy jobs; leave the dead leaves 🍂 | Test, iterate, deploy! 🚀 Measure → iterate → … |
| 2026-09-23 17:37:12 | 5878 | 3180a59 |
215/367 (215.gif) | schedule/evgeniimatveev | 📡 SHIP SMALL, SHIP OFTEN • RUN #5878 — Prune Legacy, Keep Essentials ✂️ | Halfway There — Keep Automating! 🛠️ Profile The Hotspots, … |
Night-mode palettes · Daily A–N theme rotation · 14 colors · Fully automated via GitHub Actions
| 📊 Data Analyst | 🔧 Data Engineer | 🤖 MLOps Engineer |
|---|---|---|
| SQL · Tableau · Power BI | PostgreSQL · DuckDB · dbt · Docker | MLflow · W&B · XGBoost · FastAPI |
| Dashboards → KPIs → Decisions | Raw Data → Pipelines → Production | Train → Track → Deploy → Monitor |
🤖 MLOPS Insight: 📡 TEST • OBSERVE • DEPLOY • RUN #5882 — Tune models, store wisdom 📦 | Prep for an MLOps-filled week! ⏳ Monitor, alert, respond — calmly 🧭 🌅
📈 Auto GitHub Insights (UTC · auto-refresh)
Daily contributions · last 30 days · self-hosted (GitHub GraphQL → SVG, refreshed daily — no third-party renderer)
