# AI Automation & Agents — n8n, Make, RAG | Pingvin > Business process automation with n8n, Make and Zapier, AI agents, RAG assistants over your own documents, and self-hosted LLMs for private data. _Canonical: https://pingvin.dev/services/ai-automation/_ --- 1. [Home](/) 2. [Services](/services/) 3. AI Automation & Agents Services # AI Automation & Agents The work your team keeps doing by hand, done by software instead. What it is Automation of repetitive business processes, plus AI agents and RAG assistants that handle the judgement calls a plain script cannot. Who it's for Teams doing the same data entry, chasing, sorting or copying every week — usually in sales, operations, finance or support. Result The process runs without a person in the loop, in your own accounts, with alerts when something fails. [Book a call](/contact/)[Get an estimate](/pricing/) ## The problem Most teams do not have an automation problem, they have a copy-paste problem. Someone re-types a lead from a form into the CRM. Someone reads every inbound email and decides who it belongs to. Someone reconciles paid invoices against a spreadsheet on Friday afternoon. None of it is hard, all of it is expensive, and every one of those handoffs is a place where things get lost. We build the software that does it instead — and we are honest about where a twenty-line script beats an AI agent. ## What's included - Business process automation on n8n, Make, Zapier or Keragon — built in your account, owned by you - Lead processing: form submissions parsed and written straight into the CRM, with no manual entry - Lead scoring and enrichment: every new lead scored by source, budget and reply speed into a ranked list - Call processing: meetings transcribed, matched to the right CRM record, escalated when they cannot be matched - CRM hygiene: scheduled scans across every record, with each owner tagged on exactly what to fix - Invoice and payment reconciliation: paid invoices flip the deal, overdue ones start a reminder chain - Personalised outreach generated from CRM context and profile data, sent and tracked - Daily pipeline digest: new leads, stage changes, stalled deals and today's calls, in one message - AI research agent that profiles a company before your first call and writes it into the CRM card - AI inbox agent that sorts by intent, drafts routine replies and flags what needs a human - Multi-agent pipelines: qualify, research, write, schedule — with a human approving the final step - RAG systems: chat with your own documents, answers with citations to the source - Vector search infrastructure: Qdrant, Pinecone, pgvector, Weaviate or Chroma - Local and self-hosted LLMs (Ollama, LM Studio) when data cannot leave your perimeter - OCR and document parsing for invoices, receipts and contracts - Industry flows we have built before: hospitality check-in and reviews, logistics tracking and route planning, retail stock and price watch, finance reconciliation, real-estate lead-to-viewing, healthcare appointment flows ## How we work on this 1. 01 ### We watch the process first Before anything is built we map what actually happens, including the exceptions your team handles from memory. Most failed automations fail on the exception, not the happy path. 2. 02 ### We build in your accounts Your n8n, your Make, your Zapier, your API keys. You can read every workflow, change it, or take it elsewhere. There is no vendor lock to us. 3. 03 ### We assume it will break Every workflow ships with error handling and a failure alert that names what broke. A silent automation is worse than no automation. 4. 04 ### We use agents only where they win An LLM is right for intent, summarisation and messy input. It is wrong for arithmetic and routing rules. We will tell you which one you have. ## Stack - n8n - Make - Zapier - Keragon - Claude - OpenAI - Model Context Protocol - LangChain - LlamaIndex - Qdrant - Pinecone - pgvector - Weaviate - Chroma - Ollama - LM Studio - OCR / document parsing ## Powered by Claude Code We use AI-assisted development as part of the production process, not as a demo. In practice that means fewer billed hours for the same delivered scope — the saving lands on your invoice, not in our margin. Our team holds Anthropic certifications in Claude Code, the Model Context Protocol, Agent Skills and the Claude API. ## Packages ### One Workflow One repetitive process, automated end to end and handed over running. Price on request Scoped at estimate #### What's included - Discovery call and process mapping for a single workflow - Build on n8n, Make or Zapier — your account, you own it - Up to three system integrations (CRM, mail, Slack/Telegram, sheets) - Error handling and failure alerts - Handover session plus a written runbook #### Not included - LLM or AI-agent logic - Custom UI - Ongoing monitoring after handover #### You receive Working automation in your account, runbook, recorded handover. [Get an estimate](/contact/) Most popular ### Automation Suite A connected set of workflows plus AI agents, run as a monthly retainer. Price on request Scoped at estimate #### What's included - Audit of your repetitive processes with a prioritised backlog - Multiple connected workflows built and released in sequence - AI agents where they beat rules: lead scoring, research, inbox triage, document extraction - RAG assistant over your own documents, answers with source citations - Monitoring, alerting and a monthly review of what ran and what failed - Changes and tuning included for the retainer term #### Not included - Third-party subscription fees (n8n cloud, LLM API, vector DB) - Migration of the CRM itself #### You receive Released workflows, agent configs, monitoring dashboard, monthly report. [Get an estimate](/contact/) ### Custom Build Multi-agent pipelines, self-hosted models, or automation inside a regulated process. Price on request Scoped at estimate #### What's included - Multi-agent pipelines with human approval gates - Self-hosted or local LLMs (Ollama, LM Studio) where data cannot leave your perimeter - Vector search infrastructure: Qdrant, Pinecone, pgvector, Weaviate or Chroma - OCR and document parsing for invoices, receipts and contracts - Architecture document and a technical audit of the result #### Not included - Hardware procurement - Compliance certification itself #### You receive Architecture document, deployed pipeline, audit report. [Get an estimate](/contact/) ## Related cases [Personaliser.io](/cases/personaliser-io/) ### Personaliser.io Automation / AI Generate and send personalized messages based on data from LinkedIn using AI Role Tech Lead Duration 2 months - Bubble - OpenAI - LinkedIn API - HTML / CSS / JS - REST API [Le Collaborateur](/cases/le-collaborateur/) ### Le Collaborateur AI A platform that allows you to upload documents and edit them using AI, create new versions Role Tech Lead Duration 2 months - Bubble - OpenAI - Custom Bubble plugin - HTML / CSS / JS - REST API [Winery Management System](/cases/winery-management-system/) ### Winery Management System SaaS Winery Management System (WMS) is a platform for optimizing wine storage and sales processes, helping its users track wine inventory, create wine lists, and manage sales transactions. Role Team Lead · Full Stack Developer Duration 5 months - Bubble - Bubble API - Algolia - ApexCharts - PDF Monkey ## Questions about this direction ### How is an AI agent different from a normal automation? A normal automation follows rules you define: if the form says X, write X into the CRM. An agent handles input that has no fixed shape — an email whose intent has to be read, a document whose layout changes, a lead that needs researching. We use rules wherever rules work, because they are cheaper, faster and easier to debug, and reach for an agent only when the input genuinely varies. ### What is a RAG system and do we need one? RAG means the model answers from your documents rather than from its training data, and cites which document it used. You need one when people repeatedly ask questions whose answers already exist somewhere — in a wiki, past tickets, contracts or product docs — but nobody can find them quickly. If your team asks the same five questions in Slack every week, that is the signal. ### Can we keep our data off third-party servers? Yes. We deploy local or self-hosted models with Ollama or LM Studio, and self-hosted vector databases, so documents never leave your infrastructure. It costs more to set up and nothing per use, which usually pays back at volume. ### Who owns the automations you build? You do. Everything is built in your own accounts with your own credentials. We hand over a runbook and a recorded walkthrough, so your team can change what we built without calling us. ## Think this is your problem? Book a call or ask for an estimate. Both are free and neither commits you to anything. [Book a call](/contact/)[All services](/services/)