KDMG Tech Contact ES

~/tech

Kristian Macias Goncalves

Admin automation · Chatbots · AI-assisted development

I automate repetitive, administrative workflows, from idea to production.

How I build AI-assisted development, with whichever tools fit each project best: I define the problem, review the result and take it to production.

See an example ↓ Write to me →

Snake · 0

Press an arrow key or tap to start.

01

My solutions

Chatbots and assistants

Telegram bots and custom GPTs that answer people and get tasks done. I come from customer support: I know which cases a bot can close and which need a person.

Telegram · Custom GPTs · LLM

Admin automation

Repetitive tasks that stop being done by hand: an event triggers the workflow and the system does the rest, with AI when text needs to be understood or written.

n8n · Webhooks · LLM

API integration

Connecting tools, even when there's no documentation: session-based auth, CSRF tokens, pagination, file uploads and retries that don't overload the server.

REST · Sessions · CSRF

Applied AI

Generative models inside real processes: prompts that return structured JSON for another system to use. Text, image editing and voice.

Gemini · Claude · ChatGPT · ElevenLabs

Data across systems

Per-project integrations that combine data from several sources, including full CRMs, so it all lives in one place and can be worked with.

APIs · CRM · Data analysis

End-to-end product

Backend, database, mobile app and deployment, all the way to production with real users.

Node.js · Supabase · Flutter

02

What I've built

SaaS product

Wardex

Multi-account SaaS built on Vinted's internal API: it orchestrates parallel sessions and automates the item listing cycle.

In production with real users · private beta

Integration
Custom HTTP client against an API with no public documentation: session-based auth, CSRF tokens, pagination, image uploads and retries with backoff so the server isn't overloaded.
LLM
Gemini writes titles and descriptions and edits images; the output is structured JSON that the backend consumes directly.
Backend
REST API in Node.js and Express on Supabase (PostgreSQL); mobile client in Flutter.
Security
AES-256-GCM encryption, JWT authentication, rate limiting and audited Row Level Security policies.
Deployment
CI/CD with GitHub Actions, deployed on Railway and Vercel.

Beta access ↗

Data analysis

Data and measurement

Pipelines that extract, normalize and cross-reference data from several sources, including CRMs, to measure results with comparable metrics.

Ingestion
Automations that capture each system's measurements and consolidate them for analysis.
Metrics
Aggregation by period: actual cost versus consumption cost, idle capacity and unit cost per user and per account.
Output
Dashboards with monthly series to compare periods and spot deviations.

Communities and sales

Telegram bots

Bots on the Telegram Bot API that process group events to manage communities, run engagement activities and generate sales leads.

Events
Handlers for reactions, replies and photo or video messages that award points to each member.
State
Per-user balance, group leaderboard and prize redemption, available through commands; private storage of photos and videos.
Conversion
Catalog browsable by category with buttons, purchase links and a language selector to capture leads.

AI chatbot

This chatbot

Conversational agent with tool calling, deployed on serverless functions: it answers from the content of this page and records contact requests. Try it here.

Architecture
Serverless Node.js functions on Vercel; a single model call per message to keep latency and cost down.
LLM
Claude Haiku via API, behind a common interface that lets you switch providers (to Gemini, for example) with an environment variable.
Grounding
System prompt restricted to the published content; user messages are treated as data, not as instructions.
Tool calling
A tool with a strict JSON schema that, after explicit confirmation, validates name, email and message and sends them by transactional email.
Anti-abuse
Per-IP and global rate limiting, temporary bans for repeat offenders, a heuristic spam filter ahead of the model and HMAC signatures on responses.
Kristian's assistant · AI

AI: it can make mistakes. Processed with Claude (Anthropic), and Kristian gets a copy of the conversation. Don't share sensitive data.

03

What I work with

Backend
Node.js · Express · REST API
Data
Supabase · PostgreSQL
Mobile
Flutter
AI
Gemini · Claude · ChatGPT · ElevenLabs · prompt engineering · coding assistants
Automation
n8n · Webhooks · Telegram bots · Custom GPTs
DevOps
Git · GitHub Actions · Railway · Vercel
04

Let's talk_

I'm looking for projects and roles in AI automation, API integration, implementation or technical support, remote or in Madrid.

tech@kristianmacias.com

or leave me a message