Maxkaizo (@Maxkaizo_) header
Maxkaizo (@Maxkaizo_) profile picture
Joined January 2008
Photo by Maxkaizo (@Maxkaizo_)Photo by Maxkaizo (@Maxkaizo_)Photo by Maxkaizo (@Maxkaizo_)Photo by Maxkaizo (@Maxkaizo_)Photo by Maxkaizo (@Maxkaizo_)Photo by Maxkaizo (@Maxkaizo_)Photo by Maxkaizo (@Maxkaizo_)Photo by Maxkaizo (@Maxkaizo_)Photo by Maxkaizo (@Maxkaizo_)Photo by Maxkaizo (@Maxkaizo_)
In my Pokémon RAG project, I’m realizing how critical evaluation is — not just at the generation stage, but across the entire system.
1
1
56
Fancy models won’t help if you don’t know why your system fails. Before improving prompts or tuning agents, measure, analyze, and iterate. That’s where real progress happens.
1
42
The Q&A dilemma Even with hybrid retrieval, some Pokémon answers still miss the point. My intuition: documents without a question-answer format make intent ambiguous.Structuring knowledge as Q&A could be key to aligning retrieval with actual user intent.
41
I compared t-SNE and UMAP on the Pokémon embeddings stored in Qdrant. t-SNE scattered everything with little global meaning. UMAP formed clear clusters aligned with document themes. A simple change in visualization can reveal how coherent your embedding space really is.
1
84
I’m now using Qdrant to power the hybrid retrieval system for my Pokémon knowledge project. It combines dense and sparse vectors Next step: testing how structure and context influence what the model retrieves.
34
Experimenting with Neo4j inside a RAG pipeline made me wonder — do graph databases really add value here? 🤔 They’re great for relationships and reasoning, but is the extra complexity worth it?
1
63
For me, the graph complements the vector DB — not replaces it. It’s useful to validate facts, not to retrieve meaning. Still, the exercise was worth it: understanding when structure helps is key for future business use cases.
25
I’ve been building a RAG system based on Pokémon data — combining Neo4j for structured knowledge and Qdrant for semantic retrieval. It’s not groundbreaking, but it’s been a great way to learn how hybrid RAGs actually work in practice.
2
59
Most of the graph db parts are inspired in this article from "Pokégraph: Gotta Graph ‘Em All!" written by @joedepeau
1
24
Beyond the Pokémon theme, the value is in understanding data flow, chunking, embeddings, and how to make retrieval truly contextual. These details are what matter when applying RAG to real business problems.
27
🧵Just built an end-to-end ML pipeline for predictive maintenance that I'm pretty excited about Went from raw telemetry data → deployed model in a fully reproducible environment Here's what I learned about MLOps fundamentals #MLops #MlFlow #Prefect #Localstack #AWS #Docker #ML
1
1
58
Target engineering was crucial: 4h gap + 2h horizon for failure prediction Prioritized recall over precision (missing a failure >> unnecessary maintenance) Business context drives everything in applied ML
1
18
Always exploring new approaches in AI/ML 🚀 Code is public if anyone wants to dive deeper or has feedback What MLOps tools are you excited about lately? Repo: github.com/Maxkaizo/pred_mai… #MachineLearning #MLOps #AI #Python
1
1
26
Looking to level up in AI agents & LLMs? Alexey Grigorev's new bootcamp is 🔥 Practical, hands-on, and focused. Starts Oct 6 (7 weeks) Info: maven.com/alexey-grigorev/fr… Scholarships: forms.gle/JfDwCX4CRaLnz8jU9 A solid opportunity—worth sharing. #AI #LLM #RAG #Agents #AIengineering
24
BRO THIS IS THE MOST UNHINGED MINUTE OF WRESTLING YOU WILL EVER SEE 😂 OH MY GOSH #WorldsCollide nitter.gq/ShootOrWork/status/196…
114
786
61
8,837
340,173