Smart Price Matching Engine (AI-Powered Retail Analytics)

Competitor price scraper (API-first/BS4) with ThreadPoolExecutor & PostgreSQL. Product matching via Transformers + RapidFuzz (95% prec). Active Learning loop with SetFit fine-tuning for semantic refinement. Django dashboard for purchasing dept, reducing manual analysis by 75%.

Agent Orchestration with LangGraph and LLMs for Business Optimization

Development of a Multiagent System powered by LangGraph and LLMs for a supermarket chain, automating tasks and improving efficiency in different areas through advanced AI orchestration

Multimodal AI Recommendation Engine & Hybrid Search

Django platform powered by pgvector and multimodal embeddings. Solves the Cold Start problem using an asymmetric Rocchio algorithm for real-time personalization. Features a Hybrid Search pipeline combining Lexical and Semantic retrieval via Reciprocal Rank Fusion (RRF).

Framework-agnostic LLM Engine (Full-Stack)

Built from scratch (No-Framework) to master low-level prompt engineering. Features a multi-provider Factory (OpenAI/Ollama/HF), decoupled MemoryStore interface, and native Function Calling. High-concurrency Async FastAPI backend optimized with ThreadPools.

Intelligent MBA (Intelligent Market Basket Analysis)

A project that uses market basket analysis to identify relationships between products to improve retail strategies, as well as provide intelligent product recommendations to customers based on their shopping patterns.

Intelligent Travel Assistant LLM: TinyLlama Fine-Tuning using LoRA and TRL

Project focused on adapting a Large Language Model (LLM), TinyLlama-1.1B, to create a specialized travel inquiry chatbot. Supervised Fine-Tuning (SFT) was implemented using the efficient training technique LoRA (PEFT) and TRL/Hugging Face libraries to achieve high response accuracy.

Classic Intent Classifier Chatbot

Classic Intent Classifier (BoW + MLP). Non-LLM NLU for controlled dialogue automation. Features WordNet Synonym Augmentation and Regex Slot Filling for high semantic robustness.

Content-based Movie Recommendation System (TF-IDF)

​Content-based Recommendation System for a streaming platform aggregation start-up. I used the Scikit-learn library, implementing TF-IDF vectorization to describe movie content according to their synopses, and cosine similarity measure to order movies according to their similarity score.

Restaurant recommendation system based on sentiment analysis

Group project analyzing the U.S. restaurant market, targeting the states of California, Florida, and Pennsylvania. Using sentiment analysis techniques on Yelp reviews and Google Maps, we predicted the growth and decline of different items and created a recommendation system for users.

AI-Text Pro: Intelligent OCR and LLM-Powered Document Analysis

Web application that uses techniques to extract text from images as well as PDFs. It includes multilingual translation, automatic classification, summary generation and a chatbot that allows users to ask questions about the extracted text, improving document analysis.

MercadoLibre Data Mining and Storage Pipeline with Automated Alerts

Implementation of data extraction workflow with Apache Airflow and MercadoLibre API. Compilation of information of the 50 most relevant items in "MLA-MICROWAVES" and storage in a database. Automation of email alerts for sales over $X.

Prediction of energy consumption in buildings

The project consisted of predicting the energy consumption of the two office buildings of CTIC, located in CTIC office buildings, located in Asturias (Gijón and Peón). This project is part of the AI4ES 2023 Datathon.

Machine learning model for predicting machine failures

This project is a practical application of machine learning for predictive maintenance. Two approaches were used: A binary classification model that predicts whether a milling machine will fail or not, and a multiclass classification model that predicts milling machine failure modes.

Analysis of Internet access in the telecommunications market in Argentina

In this project, a complete analysis of Internet access in Argentina was carried out. This analysis allowed us to recognize trends and patterns in market behavior, and to identify opportunities for growth and improvement at the national level.

Lexi: Conversational Assistant with LLM

An LLM-powered chatbot using Ollama. Offers seamless support within web apps by understanding and responding to user questions in natural language.