Recommendation · MoneyPush permission filtering into the vector storage layer — as data is adjusted, tag it into the vector stores — because with too many documents you need to filter.
AI Engineer · Grep or Embeddings? Agentic Search Over Company Documents — George He, LlamaIndex
Recommendation · MoneyIn your pre-parsing flow, generate and persist screenshots as pages are rendered from the document.
AI Engineer · Grep or Embeddings? Agentic Search Over Company Documents — George He, LlamaIndex
Recommendation · MoneyGiving the agent a pre-built screenshot index to search through is also very useful when performing RAG on multimodal documents.
AI Engineer · Grep or Embeddings? Agentic Search Over Company Documents — George He, LlamaIndex
Recommendation · MoneyWhen evaluating an OCR pipeline (e.g., Textract, Tesseract), check whether it captures the table representation and how faithful it is to the actual content layout. If it gets these wrong, the LLM's semantic reading downstream will pay for it — it might be fine at the start but falls apart when the model lacks spatial reasoning.
AI Engineer · Grep or Embeddings? Agentic Search Over Company Documents — George He, LlamaIndex
Recommendation · MoneyTake the top 100 retrieval results and re-rank them using an LLM or a deeper process.
AI Engineer · Grep or Embeddings? Agentic Search Over Company Documents — George He, LlamaIndex
Recommendation · MoneyTo grep for specific text in a file, issue a grep command (e.g., with a regex) and then read from that specific command with a max length (e.g., pull in the next thousand characters).
AI Engineer · Grep or Embeddings? Agentic Search Over Company Documents — George He, LlamaIndex
Recommendation · MoneyIf you have 100 to 1,000 files, it may be appropriate to stick with local file search because agents can manage that in context without burning too many tokens.
hedged (may be appropriate)
AI Engineer · Grep or Embeddings? Agentic Search Over Company Documents — George He, LlamaIndex
Recommendation · MoneyBen recommends using PlanetScale's traffic control system to categorize and tag all traffic coming into your database, which is the foundation for segmenting and managing load.
AI Engineer · Move Fast and Don't Break Things: Scaling Databases for the AI Era — PlanetScale
Recommendation · MoneyAI agents are excellent at working with text files and JSON files. You can give your agent your schema and ask it to help prototype a good vschema when struggling with database scalability. On a platform like PlanetScale, once the agent has designed a vschema, you plug it in and make a deploy, and the platform gives you an interface for creating new sharded systems and scaling up your database. A lot of sharding and scaling processes can happen with the help of AI or letting AI take some control, given PlanetScale's API and command line interface.
AI Engineer · Move Fast and Don't Break Things: Scaling Databases for the AI Era — PlanetScale
Recommendation · MoneyBen recommends exposing database operations through both an MCP server and command line interfaces, so both humans and AI agents can operate the database — which he says makes for a great operating experience for both parties.
AI Engineer · Move Fast and Don't Break Things: Scaling Databases for the AI Era — PlanetScale
Recommendation · MoneyBen recommends giving an AI agent your schema and describing your scalability struggles, and asking it to help prototype a good Vschema.
AI Engineer · Move Fast and Don't Break Things: Scaling Databases for the AI Era — PlanetScale
Recommendation · MoneyBen recommends creating resource budgets that say a given segment of traffic cannot exceed a certain amount of CPU usage or backend processes, so no single segment can starve the rest.
AI Engineer · Move Fast and Don't Break Things: Scaling Databases for the AI Era — PlanetScale
Recommendation · MoneyThe speaker implemented speculative decoding with vLLM, the serving engine, and recommends it because it has really good documentation on how to get started.
AI Engineer · Is Speculative Decoding Worth It? Profiling vLLM on NVIDIA Blackwell — Akamai
Recommendation · MoneyRoute telemetry and observability alerts into a collaboration tool like Slack or Microsoft Teams, so an AI agent can automatically pick them up and kickstart an investigation.
AI Engineer · The 6 Pillars of an Agentic Harness for Production — Varun Krovvidi, Resolve AI
Recommendation · MoneyDefine very precise tool calls — log queries, metrics, dashboards, code queries — so you don't burn through tokens very quickly when making multiple tool calls. Imprecise tool calls burn through tokens very quickly.
AI Engineer · The 6 Pillars of an Agentic Harness for Production — Varun Krovvidi, Resolve AI
Recommendation · MoneyThe source recommends building eval platform read patterns that support both large-volume aggregation and in-time data snapshots, choosing an underlying database that supports both.
AI Engineer · Why Building an Eval Platform Is Harder Than It Looks — Braintrust
Recommendation · MoneyThe source describes building RBAC into the system to manage controls and permissions, and implementing data masking, as features an eval platform needs.
AI Engineer · Why Building an Eval Platform Is Harder Than It Looks — Braintrust
Recommendation · MoneyMarc Klingen recommends using a coding agent pointed at the Langfuse agent skills to ask what has happened in an application and how it can be improved; he notes this would also generally work with other systems, though Langfuse is well positioned for it.
AI Engineer · The Self-Improving OSS Agent Stack — Marc Klingen, Langfuse
Recommendation · MindHuberman advises that when a task is not damaging to self or others, you should do it even if you loathe it, just for the sake of doing it. When you finish, acknowledge it — 'all right, got it done, I'm just glad to have it done' — and notice that the reward lies in the completion of the effort itself.
Andrew Huberman · How to Improve Motivation & Overcome Procrastination | Dr. Masud Husain
Recommendation · MindDr. Husain recommends that when giving motivational messages to someone pursuing something hard, say: 'This is a worthwhile pursuit. It's hard. Keep going. Trust me. It's going to work out in the end, even if it doesn't work out the way you think.' He notes these are very valuable messages to hear.
Andrew Huberman · How to Improve Motivation & Overcome Procrastination | Dr. Masud Husain