Recent AI Solutions
Helping organizations overcome challenges to deploy & scale AI
SFL Scientific Powering CBS Sports Analytics
From next generation stats and fantasy to play-by-play insights, analyzing millions of data points helps dissect the game matchups and enhances the fan experience. ViacomCBS leveraged the expertise of SFL Scientific to create in-game probability models, utilizing real-time data analysis and machine learning to create custom win and play by play analytics for each major sport.
Capturing data across traditional box scores, play-by-play, historical statistics and game conditions, and utilizing 100s of new metrics, SFL Scientific built machine learning solutions giving CBS unique capabilities for real-time score prediction, stats, and win probability predictions. New models give fans the most up to date Real-Time Win Probability across the CBS Sports ecosystem.


AI-Based Diagnostic Tools for Medical Imaging
Our work with InformAI focused on developing a standalone system to detect 23 conditions in the paranasal sinuses for ENT & radiology applications at human accuracy for FDA approval.
The radiology tool displays the indication, location, & a prediction heatmap, processing 3D CTs images, acting as an end-to-end diagnostic pipeline that ingests real-time scans and is currently being utilized by radiologists in the Texas Medical Center.

AI Inspection & Defect Detection Using Drones
Our work with a national power organization included developing and deploying a multi-label classifier to detect various defects such as corrosion, cracks, honeycombing, and other defects in large industrial structures.
The system is capable of recognizing desired features in real-time drone images and video, allowing inspection to occur at the edge or saved for later batch processing. The deep learning-based system has segmentation accuracy in the mid to high 90s.

Natural Language Processing for Petabyte-Scale Data Monitoring
Our work as a part of a larger US Army effort focused on cutting edge NLP solutions leveraging BERT and custom language models to develop a question and answer system for situational awareness and intelligence.
The deep learning powered search engine finds top relevant passages from billions of multilingual documents and allows for entity extraction, topic modeling, location analytics, and other key features. The solution is designed to work in any environment, on-premise, via AWS, or local services, and scalable to output results in seconds.

We help identify, create, and launch AI products, tools, or services that drive business. See how we can help your R&D initiatives.
Predicting Top Charitable Donors
Donor contributions are a significant part of success for the American Heart Association (AHA). SFL Scientific developed an unsupervised machine learning model to segment its customer base, profile existing donors, and discover characteristics of high-value donors and their utilization across AHA’s complex network. SFL Scientific’s data science consulting team was able to unlock highly specific demographic and geographic correlations which helped AHA build targeting lists, marketing campaigns, and manage their personnel and network development resources more efficiently. We help clients unlock growth opportunities across all marketing and commercial levers, no matter what industry. This includes reinforcing digital channels for patient or clinician engagement and deploying AI to predict purchase behavior and define next steps for sales representatives.
Learn more about our work for Healthcare & Life Sciences >
Cancer Diagnosis from Blood Samples
Biodesix develops non-invasive, blood-based and liquid biopsy diagnostic tests for oncology. We designed an AI capability to analyze data from custom LC-MS/MS and Dynamic Light Scattering (DLS) data to determine whether patient samples tested positive for certain types of lung cancer. SFL Scientific was given a very limited patient dataset comprised of patient samples and frequency spectrum readings. SFL Scientific implemented a time series classification algorithm to predict cancer from the DLS readings and achieved a 97% F-measure score against an independent, blinded, hold-out dataset. Further work included a instrumentation and cross-tool calibration model that replaced outdated development tools during production.
Learn more about our services in Healthcare & Life Sciences & Pharmaceuticals >
AI Solutions for Field-Level
Crop Metrics & Agriculture
Farmers Edge is the leader in precision agriculture. They provide advanced agronomic solutions to identify and map field variability, optimizing crop inputs that result in higher yields, better quality, and less environmental impact. Farmers Edge worked with SFL Scientific to provide customers with crop forecasting. Our team built a forecasting solution to predict import crop metrics across millions of acres by employing deep learning models and state-of-the-art pixel intensity convolutional neural networks in conjunction with standard growth and vegetative index features to predict the field-level crop yields. SFL Scientific’s data scientists are experts in working with complex and multi-modal image data and developing solutions for GIS and satellite image analysis applications.
AI Tools for European Regulatory
Compliance & Pharmacovigilance
Our work with Cunesoft used natural language processing to design an information extraction engine that automatically logged and tracked key medical terms and dependencies from clinical documents in over ten European languages. The system ingested raw regulatory and information pamphlets, such as Patient Information Leaflet (PIL) and Summary of Product Characteristics (SmPC) information, with extreme accuracy to populate relevant downstream processes. The information extraction models work on medical pamphlets, leaflets, and documentation that automatically parses pertinent key information e.g., drug name, dosage amounts, comorbidities, etc. for automating regulatory and safety compliance. SFL Scientific created ensemble models of feature-based class predictions and several sequence labeling methods, and deployed the system via Microsoft Azure. SFL Scientific wrapped the model in a web API for easy access, upload, and management as well as future scaling.
Learn more about custom NLP & automation tools
for Healthcare and Pharmaceuticals >
L’Oréal’s R&D and product innovation division wished to develop a software framework to classify acne severity using user-uploaded photographs. SFL Scientific developed a modern classification solution via CNNs to tackle the inherent variability in the skin and image data, as well as to handle real-world constraints. SFL obtained high-match accuracy for predicting the acne severity when comparing to trained dermatologists. As acne is not only a physical condition, where studies have linked acne outbreaks to emotional conditions such as depression, anxiety, and low self-esteem, the ability to recommend treatments and engage customers, beyond a retail environment, in real-time, is profound. By using an image recognition algorithm to diagnose the severity of acne, L’Oréal can extend their recommendation engine, create solutions for quick diagnosis, and aid their customers without direct access to a dermatologist for making better decisions. Learn more about how we work with retail and product organizations >
State of the Art 3D Modeling
Our work with Verizon involves working with R&D teams to create novel applications using computer vision and pattern recognition research to create various systems for high-resolution facial and movement analysis and rendering. The continued advancement in CNNs and GANs allows for accurate modeling of 2D objects into 3D shapes to generate synthetic and life-like images for VR/AR-type environments.
Solving AI R&D Challenges






Automated Clinical Data Review using Graph Databases
Working with a global pharmaceutical contract research & trial management organization, SFL Scientific developed an automated data review system combining network analysis anomaly detection and AI data review. The solution allows of richer representations of relational and traditional non-connected clinical, patient, and temporal information, allowing for automated data review and holistic examination of outlier and atypical medical patterns.
Detection Ability: The solution is built to support both traditional on-prem graph (Neo4J, Orient, etc.) and cloud-based services (AWS Neptune) and allows trial managers to monitor and examine the potentially thousands of exceptions such as adverse events to reduce manual review queries. With latent vector and graph analysis coupled with ML for data review, the novel capability provides clustering and insights into patient ‘communities’ by custom data variables (AE, Lab, Demo, History, etc.).

Predicting Sales & Inventory
Our work with an international clothing company with over 3,000 locations and over 50,000 third-party retailers and mass merchants was to develop algorithms and AWS-based infrastructure to predict sales and sell-out events for men’s and women’s new products 12-months in advance using historical sales and product metadata.
Detection Ability: Our data science consulting team exposed problems in data quality and reporting methods, cleaning up internal systems and unifying base ARIMA models with boosted predictors & features to improve demand, supply chain, and manufacturing operations, where the inclusion of historical data showed substantial improvements over previous models, with over 85% accuracy across all 40 product groups (1,600 unique product SKUs).
Learn more about our solutions for Retail & CPG >
