GRUPO PAZ
Book a call
Nearshore software & applied AI · El Salvador · US Central Time

Custom software, powered by AI.

We build web systems, data platforms and AI agents that run in production, not in a slide deck.
And we know when AI is not the answer.

Book a call See our work
01 — What we do

Six disciplines, one technical team.

From the system your staff uses every day to the data, cloud and AI behind it. Hire us for one piece or for the whole project.

Custom web systems

Internal platforms, portals and APIs built around your actual process, not a template.

LaravelLivewirePython · FastAPIReactMySQL · PostgreSQL

Applied AI

Agents that carry out tasks, RAG assistants that answer from your own documents, chatbots and automation of repetitive work.

LangGraphLangChainRAGWorkflows

Data

We centralize your information in a data warehouse and turn it into dashboards that update themselves.

Data warehouseDashboardsPower BI

M&E platforms for nonprofits

Platforms to track participants, activities and indicators, and report to donors without hand-built spreadsheets.

IndicatorsField formsDonor reporting

QA, audits and modernization

We review existing systems, document the risks and migrate them in stages without stopping operations.

TestingCode auditsLegacy systems

AWS cloud & infrastructure

Servers, databases, backups and monitoring set up and managed on AWS.

EC2S3BackupsMonitoring
02 — AI in production

Not demos. Workflows that already run.

Four ways we use AI inside real systems, each with a different architecture. For each one, we explain why we chose it over the alternatives.

01 / AGENT

Triage agent with tools

Takes a request, calls its tools and decides over several steps: resolve, ask for more information, or escalate to a person.

WHY AN AGENT?Each request needs different steps; a fixed workflow can't cover every case.

InputRequestEmail, form or chat
AgentClassifies intentWhat is asked, how urgent
ToolsLooks upDatabase, history and rules
DecisionNext step
ResolveAsk for infoEscalate
LogCase recordedTraceable and auditable
LangGraphTool callingHuman-in-the-loopStep limits
02 / RAG

An assistant that answers from your documents

Answers questions using only your internal documents —manuals, policies, reports— and cites where each answer came from.

WHY RAG?Documents change; with RAG you just re-index them, no model retraining.

QuestionIn plain language“What's our travel policy?”
RetrievalHybrid searchSemantic + keyword
ContextRelevant passagesEach with its source
ModelWritesOnly from retrieved context
OutputAnswer with citationsVerifiable in the original
EmbeddingsVector storeCitationsEvals
03 / WORKFLOW

Automated marketing content

Drafts posts and emails from a topic, following your tone, audience and brand guidelines.

WHY NOT AN AGENT?The steps never change: a workflow with one model call is cheaper and more predictable.

InputTopic or productWhat you want to say
RulesBrand guideTone, audience, channels
GenerationDrafts per channelStructured output
Human reviewEdits and approvesNothing ships unapproved
OutputPublishingScheduled per channel
WorkflowStructured outputVersioned prompts
04 / SCHEDULED

Scheduled research reports

An agent researches sources on a fixed schedule, filters what matters and delivers a ready-to-read report.

WHY THIS SHAPE?The question is the same every week; only the sources change. A scheduler sets the timing, the agent does the research.

ScheduleFixed timinge.g. every Monday morning
AgentResearchesWeb sources and internal data
SynthesisFilters and summarizesLinked to every source
DeliveryReportPDF or email
TeamDecidesAI informs, people decide
SchedulerWeb searchResearch agent
03 — Our approach

Not every problem needs an agent.

We pick the simplest solution that solves the problem well, and only move up a level when the previous one falls short. The result is cheaper, more predictable and easier to maintain.

LEVEL 1

Code & rules

When the rules are clear and stable.

Tax calculations, ID validation, permissions.

No AILaravel · Python
LEVEL 2

Orchestrated workflow

When the steps are known and run in a fixed order.

Approvals, data loads, scheduled reports.

QueuesCronOrchestrator
LEVEL 3

A single LLM call

When there's text to interpret, but it's a single task.

Classify an email, extract fields from an invoice.

Structured outputValidated JSON
LEVEL 4

RAG

When the answer lives in your own documents.

Questions about policies, manuals or contracts.

EmbeddingsHybrid searchCitations
LEVEL 5

Agent

When the next step depends on what it finds along the way.

Triage with tools, multi-source research.

LangGraphState & limitsHuman-in-the-loop
OBSERVABILITY & EVALUATION

In production, AI gets measured.

TRACESWhat the model received, which tool it called and what it answered, step by step.
EVALSTest sets that run before any prompt or model change.
COST & LATENCYTracked per request, with alerts when they drift.
GUARDRAILSOutput validation, step limits and escalation to a person.
LangSmithLangfuseOpenTelemetry
STACK

Proven tools, chosen case by case.

BACKEND
Laravel · PHPPythonFastAPI
FRONTEND
LivewireReact
AI
LangChainLangGraphOpenAIAnthropicOpen models
DATA
MySQLPostgreSQLVector storesPower BI
CLOUD
AWSQueues & workflows

We know the tools. What we offer is knowing when to use them, and when not to.

04 — Our products

DTE360

E-invoicing SaaS for El Salvador.

Issues every type of electronic tax document, signed and sent straight to the Ministry of Finance. We built it, we run it and we keep improving it.

Bulk import
Upload sales from Excel or CSV, validate them and issue documents in batches.
VAT books ready to file
Business sales, consumer sales and purchases, built from the month's documents.
Signing and filing
Multi-company, cloud-based, nothing to install.
Visit DTE360
dte360.com
Libros de IVA
Ventas a contribuyentes · Noviembre
CSVPDF ↓
DOCUMENTOS
1,284
GRAVADO $
64,712.40
IVA DÉBITO $
8,412.61
FechaTipoN.º controlClienteGravadoIVA
18/11CCF…P001-000412Cliente A, S.A. de C.V.1,250.00162.50
18/11CCF…P001-000413Cliente B, S.A. de C.V.380.0049.40
19/11NC…P001-000021Cliente A, S.A. de C.V.-120.00-15.60
19/11CCF…P001-000414Cliente C, S.A. de C.V.2,040.00265.20
SAMPLE DATA · REPLACE WITH REAL SCREENSHOT
OUR PRODUCT · 02Sports

FirstRally

Tennis progress tracking and training journal.

Players and coaches log sessions, matches and goals, and see how they improve over time. It is also our lab: we add ML, AI and IoT in stages, each one only once there is data to justify it.

Visit FirstRally firstrally.app
TodayLogging & journalSessions, matches, goals and progress
Stage 2 · MLPatternsWhich training habits relate to better results
Stage 3 · AIRecommendationsTraining suggestions based on history
Stage 4 · IoTSensorsPlay data captured automatically
Stages 2 to 4 are on the roadmap. [confirm dates or status]
05 — Selected work

Real problems, systems still in use.

Four cases across four sectors. [Client names pending]

EducationP / 01

Education program [client]

PROBLEMStudent progress was tracked on paper and reached decision-makers too late. SOLUTIONA web platform to capture assessments in the field and see progress by school and grade. RESULT[measurable result pending]
HealthP / 02

Psychosocial care [client]

PROBLEMTeams needed to record sessions with sensitive information without exposing personal data. SOLUTIONA system with roles and permissions, per-participant records and anonymized aggregate reports. RESULT[measurable result pending]
Volunteering & youthP / 03

Volunteer platform [client]

PROBLEMCalls, sign-ups and volunteer hours scattered across forms and spreadsheets. SOLUTIONOne system to post opportunities, enroll young people and certify hours. RESULT[measurable result pending]
GovernmentP / 04

Public institution [client]

PROBLEMWarehouse and fixed assets tracked by hand, hard to audit. SOLUTIONA warehouse and fixed-asset system with per-transaction traceability, hosted on AWS. RESULT[measurable result pending]
06 — How we work

Faster, because AI works for us too.

We use AI in our own process to write code, tests and documentation. An engineer reviews every delivery before it reaches you.

01

Discovery

We map the process, data and constraints, and choose the right level of solution. Scope and success criteria go in writing.

02

Design

Architecture, a clickable prototype and a delivery plan before any production code.

03

AI-assisted build

Short, demoable iterations. AI speeds up the work; a person reviews and approves every change.

04

Launch & support

Go-live, training, documentation and support. Your team stays in control.

Human review on every delivery. AI gives us speed; responsibility for what we ship stays with us.

07 — Why GRUPO PAZ

A formal company with real engineering craft.

COMPANY FACTS
LEGAL NAMEGRUPO PAZ, S.A.S. de C.V. NRC[number] COMPRASALRegistered supplier COUNTRYEl Salvador TIME ZONEGMT−6 (US Central)

Nearshore for US teams

Same hours as US Central Time, USD as our currency, and we work in English and Spanish.

International nonprofits and public sector

We know donor, audit and public procurement requirements, and design systems to meet them.

Incorporated, registered government supplier

GRUPO PAZ, S.A.S. de C.V.: formal contracts and invoicing, registered for Salvadoran government procurement.

Documentation and handover

Every project ships with technical docs, user guides and team training.

08 — Contact

Let's talk about your project.

Tell us what you need. We'll reply with a concrete next step, not a generic quote.

WHATSAPP+503 7729-0437 PHONE+503 7729-0437 EMAIL[email protected]
LOCATIONEl Salvador · GMT−6
Download company profile (PDF) [PDF link]