Searching in Google is cool. But have you seen what's happening in @fqsourceAI? Check this query -> Apples grading and sorting: https://t.co/HTlsl0xBWY Do you realize how industrial buyers are searching now specialized suppliers?

@fqsourceAI Buyers are already searching the new way. Sources:

@fqsourceAI AI Agents to translate real engineering needs into supplier matches. Enriched, verified supplier profiles to boost precision. Suppliers own their data. RFQ (request for quotation) orchestration end-to-end (NDAs, structured specs, technical-economic comparisons).

Our take at @fqsourceAI We’re building https://t.co/fhJOCOqKQf to verticalize this thesis for industrial procurement, with deep knowledge and own taxonomy (our first vertical is Machine Vision)

The edge is shifting from “sourcing features” to supplier graphs + data quality at scale. Coupling Scoutbee’s discovery layer with Coupa’s spend platform should lift match rates and cut time-to-source. Clear signal in race vs. SAP Ariba for the dominant procurement network.

Coupa acquires Scoutbee — the AI race in supplier discovery is on Smart move by Coupa: acquiring Scoutbee to fold AI powered supplier search, a rich supplier dataset, and collaboration tools into a network that already connects 10M+ buyers and suppliers.

Every year, billion-dollar industrial companies waste tens of millions of euros through broken procurement processes. Procurement teams are drowning in data chaos, managing thousands of suppliers with legacy tools that can't keep pace. Discover https://t.co/fhJOCOqKQf

we're building cluely of angel investing. apply now

Con los #LLMs, los motores de búsqueda están evolucionando. Pero si Booking o Airbnb integran LLMs propios y verticalizados… ¿quién gana? Probé el de https://t.co/km4sNkAFKN y la precisión fue lógicamente mayor. ¿Dónde buscarías tú un hotel hoy?

Finding specialized industrial suppliers is slow, costly, and frustrating—hours wasted on outdated info and unclear results. We built @fqsourceAI, the first AI-powered sourcing engine for industrial solutions. We own the data, you own the experience. Trusted and quick.

Worth reading https://t.co/ZHG36QTN6E

Coming soon!! Next week we'll present v1 of @fqsourceAI supplier discovery platform for the industry. First vertical | #MachineVision. A 1,000 suppliers on it! Which new vertical should we open next?

https://t.co/7Kdrs2EDKl

Today we launched a new product called ChatGPT Agent. Agent represents a new level of capability for AI systems and can accomplish some remarkable, complex tasks for you using its own computer. It combines the spirit of Deep Research and Operator, but is more powerful than that

In short: top-k < N → not enough good hits in the first page. cosine < 0.80 → even the best hit looks weak.

Cosine < 0.80 Cosine similarity ranges from –1 to 1. A score ≥ 0.80 typically means “high semantic overlap.” If your best hit comes under 0.80, the smaller model hasn’t mapped the query close enough to any chunk in the index, so you re-embed the query with the larger model.

Top-k < N When you retrieve the k most similar vectors (e.g., k = 10) you count how many of those actually pass your relevance filter. If that count is below a threshold N (say 3), you decide the retrieval is weak and switch to the stronger embedding model.

We are changing right now our embeddings model: default to 3-small, auto-escalate to 3-large only when top-k < N or cosine < 0.80

Quick question for the #RAG startups: which embedding model are you using: text-embedding-3-small, 3-large, or something else entirely?

The app is built with Lovable + Supabase + Agent till now. My question is: should we look for a hosting to deploy the app before launching v1? We will do the migration for sure in the future. But it's good idea to do it now? 3 weeks before launching?