Portfolio

HayStack

HayStack is an open-source platform that uses AI and satellite data to map financial assets to real-world locations, enabling more accurate climate risk and ESG assessments. By reducing reliance on self-reported data, it enhances transparency and supports compliance with EU sustainability regulations.

LuLarge: Impact Data Search Engine

LuLarge is an AI-powered platform that simplifies ESG data discovery and integration across fragmented sources. Designed for GSSS bond issuers, investors, and sustainability professionals, it streamlines reporting, enhances compliance, and delivers actionable insights—empowering informed, impact-driven finance.

⁠

Open link

⁠

Space News: EO Market Insights: An AI-Driven Intelligence Platform for the Space Industry

EO Market Insights is an AI-powered platform that automates market monitoring for the Earth Observation and space sectors. It delivers curated, real-time insights on finance, policy, technology, and commercial trends—reducing research time and supporting smarter decisions for professionals, investors, and policymakers.

GeoLLM: Multimodal RAG for Satellite Data and Text

This project develops a next-gen retrieval-augmented generation (RAG) system that combines visual transformers, spectral image data, and natural language to simplify satellite image search and analysis. It enables text-based queries and automated image descriptions—cutting costs, boosting accessibility, and opening new use cases in climate, disaster response, and land-use planning.

HALP: Intelligent Calendar Synchronization Across Multiple Providers

HALP is an AI-powered calendar management system that seamlessly integrates and synchronizes events across multiple calendar platforms including Azure, Office 365, GMX, and Gmail. The system employs advanced natural language processing to understand scheduling preferences, detect conflicts, and intelligently organize appointments without manual intervention. HALP automatically categorizes events, suggests optimal scheduling solutions, and maintains consistency across all calendar services through secure API integrations. This solution eliminates the fragmentation and redundancy typically experienced when managing multiple calendars, providing a unified scheduling experience while respecting the unique features of each provider's ecosystem.

Mini ERP: M3RP

m3RP is a lightweight ERP solution tailored for micro and small enterprises. Combining no-code tech, large language models, and smart OCR, it automates bookkeeping and invoice processing—reducing admin work and improving financial oversight. Designed for scalability and ease of use, m3RP offers a cost-effective alternative to complex ERP systems, helping small businesses stay compliant and focused on growth.

⁠

Open link

⁠

Custom Vision Solutions: YOLOv & TensorFlow Expertise

Delivering specialized computer vision solutions through expert implementation of YOLOv and TensorFlow frameworks. With over a decade of industry experience, we develop tailored visual recognition systems ranging from lightweight mobile deployments to robust cloud-based architectures. Our technical consultation services guide clients through the entire process—from concept and data preparation to model training, optimization, and deployment—ensuring precise solutions for unique visual recognition challenges across industries.

Tree Mark Scanner: YOLOv-Based Mobile System for Real-Time Forest Product Traceability

This paper presents Tree Mark Scanner, an innovative mobile application that leverages YOLOv (You Only Look Once) object detection architecture to instantly recognize and decode engraved identification marks on tree trunks. Trained on a comprehensive augmented dataset of diverse trunk specimens with various marking styles, environmental conditions, and degradation states, our model achieves high accuracy in real-time mark detection. The neural network was successfully deployed to iOS devices through Apple's CoreML framework, enabling in-field scanning without requiring internet connectivity. This system addresses critical challenges in global timber tracking and supply chain management by providing an efficient, accessible solution for authenticating wood provenance, combating illegal logging, and supporting sustainable forestry certification programs. Field-testing demonstrates the application's robust performance across different forest types, lighting conditions, and marking age, offering a practical tool for forestry professionals, regulatory agencies, and timber industry stakeholders.

Tree Quality Classifier: A Machine Learning Approach for Assessing Forest Health and Timber Value

This paper introduces a deep learning system designed to evaluate tree quality based on visual inspection of trunk images. Utilizing TensorFlow and computer vision techniques, our model assesses key quality indicators including structural integrity, disease presence, growth characteristics, and commercial timber potential. Trained on an extensive dataset of annotated trunk photographs spanning various health conditions and quality grades, the classifier provides rapid, objective assessments that closely align with expert evaluations. The system demonstrates robust performance across diverse forest environments and tree species, offering valuable decision support for forestry professionals, conservation efforts, and timber industry applications. This automated approach addresses the growing need for efficient, large-scale forest quality monitoring while reducing reliance on time-intensive manual inspections, potentially transforming resources.

Tree Type Classifier: A Deep Learning Approach for Identifying Tree Species from Trunk Images

This paper presents a novel tree type classification system built using TensorFlow that can accurately identify tree species based solely on trunk images. Leveraging convolutional neural networks, our model was trained on a diverse dataset of tree trunk photographs capturing various bark textures, patterns, and structural characteristics. The classifier achieves significant accuracy in distinguishing between common tree species even in challenging conditions such as varying lighting, seasonal changes, and different growth stages. This technology offers practical applications in forestry management, biodiversity research, educational tools, and citizen science initiatives. Our approach demonstrates how specialized computer vision models can effectively address taxonomic challenges in dendrology that traditionally required expert knowledge and close examination of multiple tree features.

AI-Based Travel Portal: Phase0

Phase0 is an innovative pre-travel planning system that leverages advanced LLM technology specifically calibrated to minimize hallucination effects in corporate travel management. This platform acts as an early-intervention assistant, engaging travelers before trip planning begins to ensure regulatory compliance and optimal decision-making. Designed to support the evolving holistic approach to travel management, Phase0 integrates proprietary Human Traveler Digital Twin technology to accurately predict behavior and generate personalized, compliant travel recommendations. The system excels in handling complex scenarios like extended project team deployments, automatically generating actionable compliance checklists for regulatory requirements such as A1 documentation and work duration limits. By providing factually accurate, practical travel guidance at the earliest planning stages, Phase0 transforms corporate travel management from a distracting administrative burden into a streamlined, compliant process aligned with both organizational policies and individual traveler preferences.

AI-Based Booking Confirmation Parser

A sophisticated natural language processing system that automatically extracts and organizes critical travel information from diverse booking confirmations. This TouristMobile feature seamlessly processes emails, PDFs, and digital documents from airlines, hotels, car rentals, and activity providers, accurately identifying reservation details regardless of format variations. The parser intelligently recognizes confirmation numbers, dates, locations, pricing, and policy information, then synchronizes these details with the user's itinerary. This technology eliminates manual data entry, reduces human error, and creates a comprehensive travel timeline with minimal user intervention, significantly enhancing the travel management experience within the TouristMobile ecosystem.

Log Anomaly Detection: Log Marmot

An intelligent system that continuously monitors application logs across TouristMobile's infrastructure to identify unusual patterns and potential issues before they impact users. Log Marmot employs unsupervised machine learning algorithms to establish baseline behavior patterns and detect deviations in real-time, automatically flagging anomalies ranging from security threats to performance bottlenecks. The system adapts to evolving application behavior, significantly reducing false positives while ensuring critical issues are promptly escalated to the appropriate teams. This proactive monitoring approach has dramatically improved system reliability, reduced diagnostic time, and enhanced the overall stability of the TouristMobile platform, enabling engineering teams to resolve potential problems before they affect the customer experience.

Flight Reliability Prediction Engine: OTF

An advanced machine learning system that analyzes historical flight data, weather patterns, airline performance metrics, and airport conditions to predict flight reliability with exceptional accuracy. Integrated into the TouristMobile platform, this OTF (On-Time Flight) engine provides travelers with real-time probability assessments of delays, cancellations, and service disruptions before booking. The system continuously refines predictions through feedback loops and seasonal pattern recognition, empowering users to make informed travel decisions based on personalized reliability scores. This technology significantly reduces traveler uncertainty and helps mitigate the financial and logistical impacts of unexpected flight disruptions.

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.