How fast can you start?

A Discovery sprint runs 2 weeks and ends with architecture + execution plan. A pilot in real production lands in 4 to 8 weeks. Full systems in 3 to 6 months. We discuss timelines case by case before committing.

How do engagements work?

Discovery is a fixed 2-week scope. Pilot and Production are scoped to a specific outcome you sign off on. Fractional AI Lead is ongoing thinking-partner work with a 3-month minimum. No open-ended hours, no permanent dependency. Details on the call.

Does my data leave my infrastructure?

Not unless you want it to. On-prem deployments, dedicated instances in your VPC, or sovereign models running in Brazil are all on the table. LGPD / GDPR aware by default.

What stack do you work with?

Whatever your case actually needs. We're stack-agnostic: any leading LLM, any vector store, any orchestration layer, any frontend, any database. If you already have infrastructure, we integrate. If you're starting from zero, we help you decide. The right stack is the one that fits your case, not the one we prefer.

And shipping to production?

Depends on your infrastructure. We work with your existing cloud, CI/CD, and observability. If you don't have those set up, we help build them. The agent you see at /demo is a worked example of a full stack from scratch, ready to reuse or adapt.

What happens after delivery?

Your call: full handoff (documented and we walk away), ongoing support (SLA + on-call), or Fractional AI Lead continuing as a senior thinking partner. No lock-in.

Software Zechim builds for Brazilian retail

Every week I talk to a retail operator who asks me the same thing: "what could I use AI for, in my case, without turning it into a two-year project?"

Here are the five kinds of software Zechim delivers well in Brazilian retail. None of them require rewriting your ERP. All of them run on top of what you already have (Bling, Tiny, Omie, VTEX, Nuvemshop, Shopify, spreadsheet, whatever) and all of them deliver results in weeks, not years.

1. Post-sale WhatsApp customer service agent

Customer sends a message: "where is my order?", "I want to exchange", "it didn't arrive". Today, a human reads it, looks up the order in the ERP, replies. Tomorrow, the agent reads the message, cross-references the order number in your ERP, checks the delivery status with the carrier, replies with the real data. If the message is ambiguous or involves a costly exchange, the agent escalates to you for approval before answering.

Real impact: 60 to 80 percent of post-sale tickets are "where is it" and "how do I return". The agent handles that alone. The human team keeps what needs judgment.

2. "Talk to your sales" - conversational panel

You open a screen and ask in natural language: "which SKUs sold most in the South last week?", "customers who bought on Black Friday but didn't come back", "margin by category this month", "pending orders above X". The agent translates the question into SQL, runs it against your warehouse, and answers with a number, a list, or a chart.

This replaces the "I need a report" queue that blocks your data team. And it gives access to information to store managers, sales managers, owners, who today don't open the BI because it's complicated. It's the pattern we prove in our /demo with a fictional Acme Store.

3. Catalog automation with AI

You launch 50 new SKUs per week. Each one needs a title, description, category, tags, image treatment, initial pricing. Today that's human hours on mechanical work. The agent receives the photos and the supplier sheet, generates the description in your tone of voice, proposes category and tags, suggests price positioning against competitors. You review instead of writing from scratch.

The catalog team stops being a bottleneck. Work becomes review, not creation.

4. Campaign monitor and media reallocation

You run ads on Meta, Google, TikTok, and marketplaces. Every Monday, someone on your team exports a report, cross-references CAC and ROAS by channel, decides where to cut and where to expand. The agent does that every day. It cross-references campaign data with product margin (not just revenue), flags a campaign that is selling low-margin products, suggests reallocation, and fires an alert when a channel degrades.

Different from a dashboard tool: the agent acts (drafts the budget change in the format the traffic manager only needs to approve).

5. Dynamic pricing with human rules

Every marketplace operation lives watching competitors. Price up, price down, new offer. The agent monitors competing SKUs, cross-references with your minimum margin and stock rules, and proposes a price adjustment. You define the limits (never below X margin, never raise more than Y percent per week), the agent operates within them.

It's not "AI decides alone" - it's an agent that does the manual work of watching competitors and escalates a decision only when it hits the limit you defined.

The common pattern

All five follow the same shape:

Integrates with data you already have (ERP, marketplace, media platform, stock system)
Does the mechanical work continuously (24/7, doesn't wait for a weekly spreadsheet)
Escalates to a human when it involves judgment, high cost, or risk of error
Auditable log of everything it did or proposed
Runs on top of the stack you already pay for (your Vercel or cloud, Anthropic, Postgres or the DB you have)

None requires an internal AI team. None turns into a permanent dependency on us.

What we don't do

We won't build your next horizontal generic "ChatGPT for retail". If you want that, there are 50 SaaS at $50/month doing it worse.
We don't replace your ERP. We integrate.
We don't sell broad "digital transformation". We win by delivering ONE thing that works in 6 weeks.

If you operate retail in Brazil and one of these five scenarios hits your day-to-day, worth 30 minutes for us to understand where it fits and how we'd scope it.