Technology On the Net

"How many customers upgraded last month?" sounds like the perfect job for an LLM. The model knows SQL, your warehouse holds the answer, and a chat box beats waiting two days for an analyst. Then the first real week happens: a query joins the wrong table and doubles revenue, another scans three years of events and blows the warehouse budget, and someone asks the assistant to "clean up the test accounts" and it writes a DELETE . Text-to-SQL works in production, but only when the model is the smallest part of the system. This guide walks through the pieces that make it safe and accurate: giving the model the right slice of the schema, running it under a role that cannot hurt anything, validating every query before it runs, and showing users enough of the reasoning that they can catch a wrong answer. Image: Vasek Frolik via Wikimedia Commons (CC BY 4.0) Why naive text-to-SQL fails The common first version pastes the whole schema into the prompt, asks for...

Social Media Management Services That Move the Needle Most agencies sell you "engagement." They talk about likes, follows, and brand awareness like those metrics pay your bills. We measure social media differently. Every post, every campaign, and every ad dollar should do one of two things: make money or build an asset that makes money later. That is the standard we hold our social media management services to. Santa Fe Capital Management is not a social-first agency. We are a conglomerate that operates across real estate, software development, blockchain, and digital marketing. Social media is one lever in a broader machine built for deal flow and expansion. We do not manage accounts for the sake of looking active. We manage them because attention is currency, and you should be collecting it. Here is what we deliver when we take on your social media management. --- 1. Strategy Built Around Revenue, Not Vanity Metrics Most social media management services report impres...

Most RAG failures blamed on "bad retrieval" start one step earlier: the query itself. Users type "what about the EU one?", "pricing??", or a three-paragraph rant with the actual question buried in the middle. Your embedding model dutifully turns that text into a vector, the vector store returns the nearest chunks, and the answer is confidently wrong because the search never had a real question to work with. Image: Jernej Furman from Slovenia via Wikimedia Commons (CC BY 2.0) Query rewriting is the layer that fixes this. Before you hit the index, a small model (or a few deterministic rules) turns the raw user message into one or more search queries that actually describe what the user wants. Done well, it lifts recall on vague and conversational questions. Done carelessly, it adds latency, burns tokens, and quietly changes the question so the system answers something the user never asked. This guide walks through the main rewriting techniques,...

Most RAG pipelines are built to add knowledge. A connector pulls new documents, a chunker splits them, an embedding model turns the chunks into vectors, and the vector store grows. Very few teams design the reverse path with the same care. Then someone deletes a confidential file from the shared drive, a customer asks you to erase their data, or legal pulls an outdated policy, and the assistant keeps quoting it for weeks. Image: mikemacmarketing / photo on flickr via Wikimedia Commons (CC BY 2.0) This guide covers how to make deletions flow through a production RAG system: from the source of truth, through the chunk and vector layers, into caches, and out of the places people forget, like evaluation sets and logs. The goal is a system where "deleted upstream" reliably means "gone from answers" within a time window you can state out loud. Why deletes are harder than inserts An insert has one obvious output: new vectors in the index. A delete has to fi...

Small Business Social Media Marketing: 9 Moves That Actually Pay Social media marketing for small business owners is either a revenue channel or a time sink. Most owners treat it like a hobby — posting when they remember, chasing likes that never turn into invoices. That's not marketing. That's performance art. At Santa Fe Capital, we don't do performance art. We do deals. "Where there is money to be made, we will go!" — and that applies to your social feed as much as it applies to our real estate and blockchain plays. If you're going to spend hours on social media, it needs to pay you back in leads, sales, and contracts. Here are nine specific moves that do exactly that. --- 1. Pick Two Platforms and Dominate Them Most small businesses fail at social media because they try to be everywhere at once. That's a strategy for burning out, not for making money. Choose two platforms where your actual buyers spend time. For most B2B and local service busines...

Business and Social Media Marketing: Nine Plays That Pay, Not Just Post Let's be direct. Most business and social media marketing advice comes from people selling courses, not closing deals. Santa Fe Capital Management is a California-based conglomerate operating across commercial real estate in Northern California, web and mobile development, blockchain, derivatives trading, and digital marketing. We treat marketing like every other line on the balance sheet: it either produces a return or it gets cut. That's the standard we hold for our own portfolio, and it's the standard we bring to the clients and partners we take on. This list is for two audiences. Business owners who need practical, no-nonsense marketing that drives inquiries. And investors who want to know whether their portfolio companies are spending marketing budgets like capital — or burning them like a hobby. If you're looking for warm-and-fuzzy branding talk, stop here. Where there is money to be made, w...

Giving an LLM agent a "run code" tool is one of the fastest ways to make it useful. It can crunch a CSV, test a regex, plot a chart, or check its own math instead of guessing. It is also one of the fastest ways to get burned. The code the model writes is untrusted input, no matter how well-behaved the model usually is, because the model's instructions can come from a user, a retrieved document, or a web page that someone else controls. Image: mikemacmarketing / photo on flickr via Wikimedia Commons (CC BY 2.0) This guide walks through how to run agent-generated code safely in production: which isolation layer to pick, which limits to set, how to handle files and network access, and what to log. The goal is simple. A bad snippet should fail inside a disposable box and take nothing else with it. Start from the threat model, not the tool Before you pick a sandbox, write down what you are defending against. For most agent setups the realistic list looks like t...