Alberto – Python, LLM, AWS

AI Smart Data Processing is Zantaz's flagship AI-driven platform designed to transform sprawling data estates—especially unstructured and “dark” data—into actionable assets. Rather than merely storing unused files, it brings enterprise data under robust governance, accelerates insights extraction, and dramatically reduces cost, risk, and compliance exposure. Data Optimization+10

• Scanned large-scale data estates—up to 8 million files per hour—without transferring data off-site;
• Automatically indexed and enriched unstructured data through content analysis, identity directory correlation, and inferred ownership;
• Applied policy-driven actions such as tiering, retention, and deletion, based on compliance, utilization, security, and internal policy requirements;
• Generated Smart Data Collections to support targeted governance and analytics without physically relocating files;
• Enabled large-scale data discovery and classification to support cost reduction, risk mitigation, and improved data utility.

A document processing system powered by Large Language Models (LLMs), designed to ingest unstructured or semi-structured documents (e.g., PDF, DOCX, TXT) and extract relevant information to produce clear, human-readable reports. The solution utilizes advanced natural language understanding to summarize key insights, detect entities, classify content, and structure data for further analysis. It supports customizable prompts and templates to enable domain-specific outputs, making it suitable for various use cases such as legal briefings, medical documentation, financial reporting, and compliance reviews.

• Implemented full-text search across large unstructured datasets, optimizing indexing and query performance;
• Built an event-driven agent to manage services asynchronously, improving integration and scalability;
• Integrated LLM-powered web search, enhancing relevance and semantic understanding of results;
• Developed a smart RAG pipeline for contextual document-grounded responses;
• Created document text extraction tools to process and structure content from various file formats (PDF, DOCX, TXT).

AI-powered messaging platform tailored for recruitment use cases. The system enabled real-time, human-like communication with candidates across SMS and other messaging channels, enhancing candidate engagement and improving outreach efficiency for hiring teams.

• Reduced message processing latency by 40% through optimization of Django REST endpoints and asynchronous Twilio webhook handling;
• Enabled full-text search across 10M+ messages by designing and deploying Elasticsearch indices with custom analyzers for recruiter-specific queries;
• Delivered a scalable multi-tenant messaging API, supporting hundreds of enterprise recruiters and thousands of concurrent candidates;
• Integrated real-time ML classification via AWS SageMaker, automating candidate engagement scoring and reducing manual triage by 60%;
• Improved recruiter productivity by 30% through automated conversation insights and smart filtering powered by Elasticsearch and NLP techniques.

Machine learning pipeline for object detection and area estimation in agricultural land, targeting key crops like guava, passion fruit, oil palm, and coffee. The solution supported precision agriculture by enhancing crop monitoring and enabling accurate land use estimation for both smallholder and commercial farms.

• Processed high-resolution orthomosaic and satellite imagery using QGIS and custom Python scripts to prepare labeled datasets with geo-referenced annotations (GeoTIFF, SHP);
• Trained and fine-tuned deep learning models (YOLOv5, Faster R-CNN) for object detection of tree crops using PyTorch. Adjusted anchor sizes and class weights for unbalanced datasets;
• Implemented post-processing routines to convert detections into geo-polygons and calculate cultivated area using PostGIS and NumPy-based geospatial math;
• Conducted field validation and precision-recall evaluation, achieving 85–92% mAP across target crop types. Iteratively improved performance with augmentation, transfer learning, and model ensembling;
• Packaged the workflow into reusable scripts and containerized inference with Docker. Integrated results into a Django-based dashboard for agronomists and government partners.