Tensor Labs

Our project focused on optimizing mineral data analysis through advanced Geographic Information Systems (GIS) techniques. We developed an Extract, Transform, Load (ETL) pipeline to efficiently clean and format incoming geospatial data on minerals. This involved meticulous handling of missing data and the removal of anomalies to ensure database consistency. Following data refinement, we stored the information in a PostGIS (PGIS) database, prioritizing data integrity and database performance. Building upon this foundation, we crafted a diverse array of machine learning models tailored for interpolation based on user preferences. These models encompassed variogram-based methodologies, kriging, k-nearest neighbors (KNN) interpolation, and inverse distance weighting (IDW), among others. Furthermore, we implemented anomaly detection algorithms to facilitate outlier removal, providing users with comprehensive data statistics. Through an intuitive interface, users could define custom polygons and apply filters based on mineral attributes. Our models seamlessly processed this input, generating dynamic heatmaps with optional contour overlays, delivered in GeoTIFF format for enhanced visualization and analysis.

Developed a Smart Contract Auditor tool that operated in three phases, showcasing our team's adeptness in blockchain security and software development. In the initial phase, we employed Generative AI to construct an AI pipeline capable of auditing ERC-20 token contracts. This involved a through examination of contract source code to pinpoint vulnerabilities such as backdoors and centralization of control. The outcome is a detailed PDF report, offering clients valuable insights into their contract's security status. Expanding our capabilities in phase two, we scrutinized decentralized applications (dApps) by utilizing Generative AI and RAG pipelines to analyze GitHub repositories. The aim was to ensure the integrity and security of dApp implementations. Clients receive comprehensive audit reports in PDF format, empowering them to make informed decisions about their project's security posture. In phase three, we enhanced our platform by integrating user referrals and Coinbase payment methods. Additionally, we incorporated Tokenomics data from CoinGecko and Coinbase into our audit reports. This comprehensive approach underscores our commitment to delivering holistic solutions that meet the evolving needs of our clients. Technologically, our foundation rests upon Node.js, React, Python and FastAPI, showcasing our dedication to employing cutting-edge tools and frameworks to deliver robust and scalable solutions. With our Smart Contract Auditor, clients can trust in our team's expertise to safeguard their blockchain projects with confidence.

Developed an AI-powered cold-calling assistant for The Finance Group Australia. The system integrates seamlessly with Zoho CRM and Twilio to automate outbound calls. The AI assistant engages prospects with predefined questions, evaluates their responses and assigns lead scores based on interest and relevance. This intelligent approach allows telemarketers to focus their efforts on high-potential customers, significantly improving efficiency and conversion rates.

Developed a Machine Learning-driven FinTech platform that automates the collection, transformation and forecasting of financial data, showcasing our team’s expertise in data engineering, applied ML and cloud-native application development. In the initial phase, we provisioned a secured environment on AWS EC2 with hardened server access, automated updates, and Amazon RDS (PostgreSQL) for structured financial storage. Data persistence and archival were managed through Amazon S3, ensuring both raw and transformed datasets were available for downstream analytics. Building upon this, we constructed Python-based pipelines to collect equities data, options contracts, and global economic indicators via Yahoo Finance and additional APIs. These pipelines perform historical backfills to 2020 and support continuous daily and weekly refresh cycles. Robust validation and error handling were integrated, with data quality reporting surfaced through the platform. Advancing to the ML stage, we integrated proprietary predictive models for call/put options and multi-day stock forecasts. Predictions leveraged a dynamic U.S. market holiday calendar and precision-driven thresholds, with Stage 1 outputs feeding into Stage 2 models for refined accuracy. Forecasts and performance metrics were persisted in Amazon RDS and automatically exported in CSV/Excel formats for end users. The system was delivered as a FastAPI backend powering automated ML workflows, coupled with a Streamlit dashboard for visualization and user interaction. This combination enabled stakeholders to monitor pipelines, review forecasts, and access historical analytics through an intuitive web interface. Technologically, our foundation rests on Python, FastAPI, Streamlit, AWS EC2, Amazon RDS and Amazon S3, with automation orchestrated through cron-based scheduling and modular ML integration. The resulting system is a production-ready platform that delivers actionable, data-driven insights, empowering financial decision-making with measurable accuracy.

The project features an intelligent assistant chatbot that seamlessly interacts with the database to retrieve data and deliver meaningful transaction insights based on user queries. Beyond simple lookups, the chatbot can generate dynamic visualizations and plots for better understanding. It also supports advanced, calculation-driven questions, providing accurate and context-aware responses to even the most complex queries.

An intelligent system designed to simplify and automate the management of Bills of Materials (BOMs). Users can provide input through emails, PDFs, Excel files, or even images, regardless of format or structure. The system automatically extracts and processes the relevant information, delivering a clean, structured BOM in JSON format. Beyond extraction, it allows users to search for specific parts or materials online and instantly retrieve detailed information. It also supports natural language inputs, simply describe your setup and the system will generate an estimated BOM after asking a few clarifying questions. Additionally, users can easily modify existing BOMs by specifying the desired changes, ensuring flexibility and precision at every step.

Developed an end-to-end crypto market intelligence platform that ingests, processes and models multi-modal financial and sentiment data in real time. The system predicts short-term market breakouts, sharp up or down price movements across major exchanges using deep feature engineering on market microstructure, derivatives, on-chain and social sentiment data. Powered by a modern stack of FastAPI, PostgreSQL, S3, Redis, and Streamlit, with LightGBM/XGBoost models and SHAP-based interpretability, the platform achieves ~75% recall and 71% precision in live breakout prediction. Our infrastructure includes automated ETL pipelines, feature stores and a real-time inference dashboard, making it a scalable foundation for quantitative crypto analytics and trading research.