[ APPLIED AI // SYSTEMS ARCHITECTURE ]

Turning data into autonomy and processes into intelligence.

I design tools to prepare information, automate tasks and apply AI to real workflows.

Focus
Data preparation
Application
Useful knowledge
Approach
Real workflows
khipuwasi_system_flow.svg
Select a node
01. SOURCES & FILES 02. CLEAN MARKDOWN 03. KNOWLEDGE BASE 04. SUPPORT & RESOLUTION
NODE_01

Sources and file formats

Multimodal input

Collect internal documents, meeting recordings, training videos, screenshots and operational guides in several formats.

PDF / DOCX Audio / Video Images / Diagrams
// MANIFESTO

We design AI for real workflows.

The Andean khipu recorded information using cords and knots. It inspires our approach to AI: structure information and verify the results.

[01] SYSTEMS FOR PRODUCTION

Production first

We design for everyday use, with response time targets, fallbacks, evaluations and cost controls.

[02] DATA TOPOLOGY

Your data is the starting point

We organize data and design retrieval so each answer can draw on relevant, verifiable information.

[03] THE WASI ETHOS

Open knowledge and technical rigor

Wasi means home. We want to share design decisions, reproducible examples and what we learn from our projects.

[ PRODUCTS & TOOLS ]

Projects in development

These tools share a premise: well-structured information helps AI produce reliable answers and saves time.

NODE_04 // SUPPORT & RESOLUTION Prototype

KhipuTR

Tucuy Ricuy Engine

A troubleshooting and technical support agent. It consults company guides, walks users through a resolution and prepares a summary for support if the issue needs escalation.

View project
Answers with citations Internal guides Escalation with context
khiputr_widget.ui
ILLUSTRATIVE EXAMPLE
CUSTOMER We rotated the production API key, and the billing webhook now returns a 401 error. Where do I update the secret?
KHIPUTR Verified in the guide

After rotating the API key, regenerate the Signing Secret for the webhook under Settings → Integrations → Webhooks and update the WEBHOOK_SECRET variable on your server.

Source: Integrations_Guide_v2.4.md · Section 3.2
NODE_02 // DATA PREPARATION Prototype

KhipuMD

Token Optimizer

Converts documents, audio, video and images into structured Markdown for use with AI.

View project
Documents, audio and video Images and screenshots Clean Markdown output
KhipuMD.exe
ILLUSTRATIVE EXAMPLE
output: operations_manual.md Structured Markdown
---
source: "operations_manual.pdf"
type: "technical_document"
---
# Order validation procedure
## 1. Automatic approval rules
- Orders under **€2,500** with confirmed stock are approved automatically.
- If the customer's balance is over 15 days overdue, send it for **Financial Review**.
// AREAS OF WORK

Four pillars of applied AI engineering

These four areas guide how we design Khipuwasi products.

Autonomous AI agents

Agent design with clear boundaries, structured tool use and verifiable execution steps.

Multi-Agent Graphs Tool-Use Contracts Guardrails

High-performance RAG systems

Hybrid search, reranking and retrieval evaluation to support answers with relevant sources.

Hybrid BM25 + Dense Cross-Encoders RAG Evals

Advanced database integration

Combining operational data with vector indexes and knowledge graphs to query multiple sources.

PostgreSQL / pgvector Knowledge Graphs CDC Pipelines

Workflow optimization

Task automation with human review, traceability and tests for critical steps.

State Machines Human-in-the-Loop OpenTelemetry
[ ABOUT KHIPUWASI ]

From sharing practical architecture on LinkedIn to building AI tools

Khipuwasi is a professional brand focused on applied AI and systems architecture. Its name brings together the khipu , the Andean system of cords and knots used to record structured information, and wasi, meaning home or workspace.

Khipuwasi started as a community on LinkedIn to share engineering decisions and practical lessons. Today, this site showcases tools that help people and companies prepare information, access internal knowledge and improve their workflows with AI.

Focus
Applied architecture
01.
Origins on LinkedIn A technical community for RAG systems, agents and applied architecture.
02.
AI-ready information Converting and organizing multimodal data so answers can draw on reliable information.
03.
Software for everyday operations Tools focused on specific operational problems.
[ CONTACT ]

Let’s talk about the projects or a possible collaboration

If you’d like to learn more about KhipuTR or KhipuMD, share feedback or explore a collaboration in applied AI and systems architecture, get in touch.

Juan Herbozo will respond directly.