AI Projects

Knowledge Retrieval Agent: AI That Only Knows What You Tell It

May 27, 2026May 9, 2026 by Carlos Paredes

Illustration of a glowing document vault with beams of light connecting folders to a central AI brain node. Mathematical vectors and floating numbers surround the connection lines.

building a RAG agent that forces an AI to answer only from your private documents, and discovered a bug.

Async Standup Summarizer: AI Team Reports Without Meetings

May 27, 2026May 8, 2026 by Carlos Paredes

Top-down desk scene showing four sticky notes arranged around a central document labeled "Team Report." Each sticky note has a few handwritten lines.

building an AI agent that reads standup updates from a file and produces a consolidated team report in seconds, no meeting required.

Connecting DeepSeek V4-Pro to Hostinger VPS Using LiteLLM and Open WebUI

May 28, 2026April 26, 2026 by Carlos Paredes

Architecture diagram showing DeepSeek V4-Pro connected through a LiteLLM proxy to a Hostinger VPS, enabling OpenAI SDK compatibility for self-hosted AI inference.

how to run DeepSeek V4-Pro on Hostinger VPS with LiteLLM and Open WebUI

AI Assistant Tuning: What Broke and Finally Worked

May 28, 2026April 25, 2026 by Carlos Paredes

Before and after comparison showing AI assistant performance metrics including hallucination rate, response latency, context retention, and relevance score improvements after systematic tuning.

Fixing a local AI voice assistant that ignored its system prompt — the Modelfile approach that worked

I Built a Private AI Voice Assistant – Runs Entirely on My Laptop

May 28, 2026April 25, 2026 by Carlos Paredes

Laptop screen illustration showing a glowing microphone icon with sound wave rings and a privacy lock badge, representing a fully offline private AI voice assistant running locally without cloud services.

Building a private self-hosted AI voice assistant with Ollama and Open WebUI on Windows.

How I Built a Google Search Console Automation With n8n

May 28, 2026April 22, 2026 by Carlos Paredes

Workflow diagram showing automated pipeline from Google Search Console data through an n8n automation node triggered by a cron schedule to generate and deliver SEO performance reports.

I built an n8n automation to ping Google Search Console daily and notify me on Telegram. Here is what broke and how I fixed it.

From Laptop to Cloud: Migrated My Discount Monitor to a VPS

May 28, 2026April 21, 2026 by Carlos Paredes

Migration diagram showing the transition of an automated discount monitoring workflow from a local laptop setup to an always-on virtual private server in the cloud.

A few days ago I built a simple but useful automation: an n8n workflow that checks the Google Cloud certifications blog every morning . It uses a local AI model to detect any exam discounts or promo codes, and sends me a Telegram message with the result. No more manually checking. No more missed deals. … Read more

I Chose the Wrong Tool

May 28, 2026April 20, 2026 by Carlos Paredes

Decision diagram showing crossed-out wrong tool choices on the left contrasted with the correct SDK selection on the right, illustrating lessons learned from poor tool selection in AI development.

I ran the benchmarks. LM Studio was faster. So I built my home AI lab on it and discovered that speed is the wrong metric when your server needs to run without you.

LM Studio vs Ollama on a 12GB Laptop GPU

May 28, 2026April 18, 2026 by Carlos Paredes

Side-by-side comparison diagram of LM Studio and Ollama local AI runtime tools, showing feature differences including interface style, API access, setup complexity, and platform support.

I have an Acer Predator Helios Neo 16S with an Intel Core Ultra 9 275HX, an NVIDIA RTX 5070 Ti with 12GB GDDR7 VRAM, and 32GB of DDR5 RAM running at 6400MHz. Ollama is already installed with Open WebUI running locally. The model I used for every single test is Qwen3.6 35B-A3B at Q4_K_M quantization … Read more

I Ran a Good Model With Decent Hardware

May 28, 2026April 18, 2026 by Carlos Paredes

Bar chart comparing performance scores of local AI models including Llama 3, Mistral, Phi-3, and Gemma running on a mid-range consumer GPU.

I saw a benchmark showing Qwen3.6 outperforming Claude Sonnet on my laptop. I didn’t believe it. So I ran it myself — mistakes, PowerShell errors, and all.

One documented agent build in your inbox, every week