Work & Projects

Featured Projects

A selection of AI agents running in production, SaaS platforms and websites that work on their own: systems that answer the phone, reply on social media, prospect for clients and publish content with nobody sitting in front of them.

AI Agent — Real-Time Voice

Featured

Voice Agent — Phone Reception & Booking

A voice agent that answers the phone and takes the booking during the call itself. Audio streams from telephony into a real-time pipeline: speech-to-text, LLM reasoning and speech synthesis chained together, with silence-based turn detection, natural barge-in and noise suppression. It records the appointment in the database and can reschedule by identifying the caller from their number. Multi-tenant: every phone number resolves its own script, voice and calendar. Google Calendar sync is built but not currently operational.

Real-time conversation with natural barge-in
Appointment recorded in the database during the call
Rescheduling by identifying the caller's number
Multi-tenant: own script, voice and calendar per number

PythonPipecatSIPDeepgramElevenLabsPostgreSQL

Private projectRead article

Comment & DM Agent — From Instagram to CRM

AI Agent — Social Media to Customers

Featured

Comment & DM Agent — From Instagram to CRM

Buying intent shows up in comments and direct messages, and that is the first thing a busy owner stops reading. This agent receives every new interaction through a webhook, classifies intent with a language model, replies publicly in the brand's voice on the right thread and — when the message looks commercial — creates a qualified opportunity in the CRM with the full conversation context. Each brand is a separate tenant, with its own voice, connected accounts and auditable interaction log.

Intent classification on every comment
Public reply in the brand's own voice
Qualified opportunity created in the CRM with context
Separate auditable log per brand

TypeScriptNext.jsSupabaseMeta Graph APIOdoo

Private projectRead article

AI Agent — B2B Prospecting

B2B LinkedIn Prospecting with a Human Signature

A prospecting pipeline that finds profiles matching the ideal customer, analyses each one, drafts a personalised message with real context… and then stops. Nothing goes out until a human approves it from Telegram: approval is a hard gate, not a notification. Only approved messages are sent, and meetings are booked through a dedicated scheduling link. Orchestrated with Temporal so no step is lost, duplicated or left half-done when something fails.

Search and analysis of matching profiles
Messages written from real context, not templates
Mandatory human approval before anything is sent
Durable orchestration: no step gets lost

TypeScriptTemporalPostgreSQLCal.comTelegram Bot API

Private projectRead article

AI Agent — Multi-Brand Content

Social Content with Scheduled Agents

Small businesses stop posting because nobody has time to write. Here there is one agent per brand and time slot: each loads the brand voice, tone rules and editorial calendar, writes the copy, generates the image and sends the finished draft to the owner's Telegram. Nothing is published without approval; approved posts are queued and delivered to Instagram, Facebook or LinkedIn at the scheduled time. A separate watchdog service reports failures even if the orchestrator is down.

One agent per brand with its own voice and calendar
Finished draft (copy + image) delivered to Telegram
Only human-approved content gets published
Failure watchdog independent from the orchestrator

Node.jsPythonMeta Graph APILinkedIn APITelegram Bot APIDocker

Private projectRead article

AI Agent — WhatsApp Support

Bilingual WhatsApp Agent

A product available for service SMBs that lose leads because messages arrive after hours: whoever replies first wins the client. The agent answers in seconds, in Spanish or English, using each brand's own persona and tone plus the last conversation turns as context, and writes every reply back to the CRM record. It is not allowed to close: when a prospect accepts a call, it flags the thread and hands the full transcript to a human. On DailyMP's own bot: 186 messages across 16 conversations between May and July 2026, with 100% of incoming messages answered.

Replies in seconds, 24/7, in two languages
Each brand's own persona and tone
Every reply written back to the CRM record
100% of messages answered on our own bot (May–Jul 2026)

PythonFastAPIWhatsApp Business Cloud APIGroqOdooSQLite

Private projectRead article

SaaS — Shifts, Clock-In & Hours

Multi-Company Shift & Time-Tracking Platform

A multi-company SaaS platform for shift planning and time tracking, in production with a lifeguarding company in Málaga. Each organisation sees only its own data, with isolation enforced in the database itself and not just in the UI. Per-service rosters, mobile clock-in, push notifications for staff, incident management and hour exports to Excel for payroll. Covered by unit tests and automated E2E tests.

Multi-company with database-level isolation
Mobile clock-in and push notifications for staff
Per-service rosters and incident management
Hour exports to Excel for payroll

Next.js 16React 19TypeScriptTailwind CSSSupabasePlaywright

Private projectRead article

Web Invoicing for SMEs, Built for VeriFactu

SaaS — Invoicing & Compliance

In progress

Web Invoicing for SMEs, Built for VeriFactu

A web invoicing package for SMEs and freelancers who must comply with Spain's RD 1007/2023 without changing how they work. It replaces the desktop program installed on a single computer with something reachable from any browser, with unlimited users and verified backups, built on a base that implements the chained invoicing record the regulation requires. First rollout in progress: submission to the Spanish tax authority has not yet been validated against their test environment, so today this is a deployment under way, not a certified system.

Chained invoicing record, as the regulation requires
Browser access, unlimited users
Verified backups, not just scheduled ones
Not yet validated against the tax authority's test environment

FacturaScriptsPHPMariaDBVeriFactuNginx

In progress

Technical Demo — Two-Agent Pipeline

Legal Analysis Engine with Confidence Scoring

An architecture and technical demonstration piece, not a client case: it is built and deployed, but with no real traffic. Two chained agents. The first drafts the case analysis for a given jurisdiction, accepting both text and PDF documents. The second independently verifies it, detects the practice area and returns a confidence score. Below the threshold the system does not answer: it escalates the case and proposes the matching specialist instead. The output is two PDFs — one for the client, one for the operator — with the full trace of what the system did.

Technical demo: no real traffic, no clients
A second agent verifying the first one
Confidence threshold deciding when NOT to answer
Client PDF plus an operator trace PDF

PythonFastAPIGoogle GeminiReportLab

Private projectRead article

Children Photography — Web Redesign + SEO

Featured Projects

Voice Agent — Phone Reception & Booking

Comment & DM Agent — From Instagram to CRM

B2B LinkedIn Prospecting with a Human Signature

Social Content with Scheduled Agents

Bilingual WhatsApp Agent

Multi-Company Shift & Time-Tracking Platform

Web Invoicing for SMEs, Built for VeriFactu

Legal Analysis Engine with Confidence Scoring

Natalia BR Children Photography

OK Car Hire

Dog Hotel Churriana

SOS Málaga

Centro Cuídate Coín

Cafetería Hermanos Portillo

Sabor Cubano Málaga

Yuri La Diferencia

Maruchy

Liudmila Reyes

Bertronit

Silvia Corpas

Abogada Irelsis

Calma Laboral

Natalia Porras Psicología

Grupo Cárdenas Rojas

UBAsesores

Mudanzas CBM

Brigada Sol

The Nina World

Cuba Tec Málaga

Gloria Bendita Fuengirola

Want your project here?