Coach Dmytro Voytko πŸ‡ΊπŸ‡¦@DmytroVoytko

Agile & Productivity Coach | Data Science Enthusiast | Google Cloud Certified Professional | Python, Data Engineering, Analysis, ML. Ukrainian πŸ‡ΊπŸ‡¦βœŠ

 Europe Asia
bit.ly/BuildBetterHabits123
Joined August 2009
6,048 Photos and videos
My favorite motivator, or mathivator πŸ˜‰ John Wooden said β—β€œIt's the little details that are vital. Little things make big things happen.” What little #changes can you do consistently to #stress less and feel more #calm and #satisfaction during the day❓ #Kaizen #Mindset
5
2
9
#Vibecoding (popular) scenario: built an app in an evening ...then spent a week learning how to make it work with #observability. My 7 lessons on #OpenTelemetry + #Prometheus/Loki/Tempo + #Grafana alerts that every AI-assisted builder needs before prod 🧡 #devops #aiengineering
3
4
72
more replies
Make your incident webhook stupid-resilient. Never fail POST /alerts on Loki/Tempo errors β€” save the alert + endpoint + dashboard link first, enrich best-effort. And short-circuit test:true pings before they pollute your incident store.
1
4
Do you use #OpenTelemetry? #Prometheus? #Grafana? Which of my 7 #DevOps gotchas resonate with your experience/challenges?
9
Lessons learned from evaluation of #LLMzoomcamp projects 1. Tested keyword vs vector vs hybrid RRF on the same ground truth with hit-rate + MRR. Vector beat keyword, hybrid edged vector. Conclusion: Ship the winner you measured, not the one you hoped for. #rag #aiengineering
1
22
more replies
5. 5 retrieval setups, paired bootstrap CIs on the difference, tie goes to simpler, winner confirmed once on held-out docs. Largest effect wasn't hybrid (+0.07), it was reranking (+0.095 MRR). Measure the stack. #rag #aiengineering
1
48
6. Answer #evaluation: 5 docs vs 3 docs tied, so ship 3 β€” cheaper. Judge model != answer model, relevance + support + deterministic retrieved, tokens logged. Cost is a metric, not an afterthought. #llm
5
3. Text search won MRR 0.858 vs hybrid 0.829 because synthetic questions echoed doc wording. Shipped hybrid anyway for paraphrase robustness β€” unproven. If your eval favors lexical, say so. Bias disclosure > winner-picking. #llm #rag
10
I just submitted Homework 1 of #AIdevtools zoomcamp 2026! Built a Household Chores Manager in #Django for small household + garden + cat 🐈🌱 Zero Django coding β†’ working app in a few steps. How? πŸ‘‡ #aiengineering #vibecoding #opencode
1
55
more replies
#Kanban Board = To do / Today / This week / Doing / Done, with ?area=cat filter + POST move actions. Generic CBVs for CRUD, one BoardView for grouping. Biggest lesson: deferring email/push to v2 saved the homework. Scoping IS engineering. What would you have cut? #aiengineering
2
32
Takeaways: - Vague prompt β†’ chaos; spec β†’ leverage - Small tasks + acceptance criteria = AI ships clean - settings.py, runserver, test β€” you can learn Django BY building Repo: github.com/dmytrovoytko/ai-d… What should I build for HW2? πŸ‘‡ #AIdevtools #agentic #coding
1
29
Trust but verify: manage.py check β†’ migrate β†’ manage.py test green (6 tests) + test-client script for board grouping / move / filter. Overdue edge cases: done+past = NOT overdue, no date = NOT overdue. Do you test models first? #testing #Django
18
Load more