Antoine Pinto

• Implemented scheduled scraping to extract relevant information from ads
• Designed a tabular view with formatting to enable effective comparison of listings
• Developed automatic detection of vehicle “points of attention” based on history
• Added advanced features including filtering and location-based search

• Developed multiple Generative AI applications tailored to different professions (HR, Analysts, etc.) and sectors.
• Deployed applications in Streamlit for testing and validation before handover to full-stack developers for production in a Generative AI portal.
• Integrated features enhancing ergonomics and user experience (cost estimation, multiprocessing, Streamlit authentication, activity tracking, etc.).
• Optimized source code to streamline usage by full-stack developers.

Examples of applications:

CV Match: Find candidates matching a need in a CV database

• Upload resumes and extract standardized information of interest: companies, school prestige, technology, etc.
• Use the filters from the extraction to sort the candidate profiles.
• Perform a natural language search and get the most interesting candidates ordered by a match score.

Knowledge Base: Interact with a documentary database.

• Create a Knowledge Base on a specific topic (contracts, press releases, etc.)
• Upload documents and define KPIs to extract from documents.
• Manage access to the Knowledge Base with multiple permission levels (Admin, Editor, Guest).
• Extract the main KPIs for each of the documents.
• Query a Knowledge Base or compare documents.
• Check the veracity of the answers by systematically generating a link to the source in the document.

Performance Report: Generate an activity report (traffic, finance, etc.) using Generative AI.

• Select the dimension of data of interest.
• Define the structure of the report (length, language, interest factor, etc.)
• Obtain causal information of the described effects using the built-in inference model
• Download the activity report in a standardized Word format to the company model.

Document Translator: Translate a document while preserving its format.

• Upload a PDF, Word or PowerPoint document.
• Offer different translation methods, each with a level of quality and pricing.
• Show translation cost estimate.

Specialized Chat Bot: Answer questions about the internal situation of the company.

• Suggest different themes: Traffic, Finance, News, Competition, Forecast.
• Obtain graphs and tables to illustrate the generated comment.

• Performed modeling at different aggregation levels and forecast scopes: 28-week per-flight forecasts and 75-week per-day forecasts
• Developed a forecast monitoring application allowing to compare estimates by business unit or by model

• Applied a customized trading algorithm to detect specific patterns in financial data
• Automated asset purchasing and resale via API using scheduled scripts on Cloud Run and Cloud Scheduler
• Stored historical trades in a database for subsequent evaluation

• AI-Powered: Implemented features for automatic creation of chatbot titles, descriptions, and logos.
• Source Tracking: Enabled source tracking to access exact document locations for AI responses
• Privacy Control: Configured privacy controls to share chatbots publicly or keep them private
• Performance Optimization: Optimized performance using ML-based response time estimation

Technical Highlights:

• Migrated frontend from Streamlit to React in 4 days using Vibe Coding
• Enhanced security with password encryption and IP whitelisting
• Deployed the application to Google Cloud Run and Streamlit Cloud
• Automated email services with AWS SES
• Managed secure document storage with Google Cloud Storage

• Custom splitting criteria implementation
• Multi-conditional node splitting
• Supports classification and regression
• Supports multiclass classification

• Superior accuracy compared to industry-standard FuzzyWuzzy (in a 1000 obs dataset)
• Easy-to-use Python package interface
• Comprehensive similarity scoring methods

• Built a database by scraping multiple sports news sources (bookmakers, statistics)
• Integrated weather data matched stadium geolocations
• Automated scraping via scheduled scripts using Cloud Functions
• Optimized investment decisions using various sports betting strategies

• Automated scraping (via Cloud Functions) of data from sports news sites
• Automated detection (via Python/pandas) of the most "exceptional" recent statistics
• Automated writing (via Generative AI) of a journalistic commentary for each statistic
• Automated generation of a graph (among pre-designed templates) for each statistic
• Automated creation of Instagram / Facebook posts
• Non-automated publication of posts on social networks

• Implemented a “divide and conquer” approach to efficiently process large datasets using clustering techniques (KMeans, CUBT, etc.)
• Applied simulations on multiple datasets to demonstrate the algorithm's robustness and efficiency
• Analyzed performance in terms of execution time and classification error, comparing results to traditional methods
• Collaborated with Aix-Marseille University to validate results and publish findings in an academic context