Sergiu Nicoara
AI Engineer · 0.94 RAG Faithfulness, Telecom-Grade AI
New to Contra
Sergiu is ready for their next project!
Which state hits hardest — IDLE, THINKING, SUCCESS, or ERROR? 👀
2 voted
18%
9 voted
82%
11 votes
Closed
A programmer completes one quiet task. One keystroke.
The system stabilizes. He rests. Morning comes.
Then the twist: 2,843,295 active users — all depending
on that one silent fix. Faithful in the least, faithful in much.
Made with Renoise Canvas: Seedance 2.0, Canvas Timeline,
piano and warm ambient audio.
#RenoiseChallenge
What does your AI agent actually do between prompt and response? Now you can see every span, every ms, every guardrail.
4 voted
67%
2 voted
33%
6 votes
Closed
If you were running LLM evaluations on your team, which feature would you use most — the metric trend charts, the top/bottom prompt variants leaderboard, or the AI-powered 'Improve this prompt' suggestion?
2 voted
25%
6 voted
75%
8 votes
Closed
Most RAG systems fail not because of the model — but because
retrieval is broken.
Dense vector search alone misses exact matches. BM25 alone misses
semantic similarity. The fix: hybrid retrieval (dense + BM25 + RRF
fusion) with cross-encoder reranking on top.
Result on real...