Soren v Solari
Verified Expert in Engineering
Soren is a master of many skillsets. With a Ph.D. in integrative neuroscience, he has delivered algorithmic patents used today by companies like Nestlé and Nissan, done business development sales at the C-level, and built scalable AWS systems. Soren has written real-time back ends, React front ends, cognitive systems, and genomic data mining algorithms from scratch. Soren is a brilliant systems thinker who can solve any problem with a rare combination of communication, architecture, and code.
Amazon Web Services (AWS), Gmail, Linux, MacOS, Slack, GitHub, WebStorm, PyCharm, React, Python
The most amazing...
...product I've built is a personalized health app, writing 100% of the Python back end in scalable AWS, 100% of the React front end, models, and everything.
- Invented and built a revolutionary personalized health application that combines medical records, nutrition, activity, and arbitrary data for users.
- Developed complex new healthcare predictive models for personalized health, including recipe and activity recommendations and blood work analysis. This model facilitates a world-class understanding of public nutrition data.
- Built a complete front-end in React using a modern approach of only functional components leveraging React hooks and including high-throughput websockets and back-end API integration.
- Deployed and managed dozens of different microservices; worked with a scalable API, websockets, DNS, and more.
- Wrote every line of the core Python microservices framework using ZeroMQ for RPC communication, discovery services, load balancing, logging, monitoring, tracing, and continuous CDI testing/deployment to AWS infrastructure.
- Worked with microservices and deployments on cloud-based infrastructure in automated ways.
- Developed web scraping technology for personal health records as well as scraping recipe information from online resources.
- Built a PDF reader model from scratch to extract information from PDFs (no third-party tools at all) starting with binary PDF input.
- Developed other core technologies around natural language processing, artificial intelligence, machine learning, and multiple other concepts integrating complex data in sophisticated ways.
- Created a chatbot from scratch with no third-party tools featuring interactive conversation, conversation tracking over time, and simultaneously/interchangeable interactivity via SMS as well as web app chatbot (through websockets).
Machine Learning | Team Lead and Architect
- Led a small team to redesign and develop a new ML infrastructure to handle multiple ML services for near real-time call center transcription applications.
- Spearheaded the development of a new text-to-text "translation" deep learning model to more efficiently produce clean transcripts, leveraging millions of transcribed calls.
- Innovated a new deep learning semantic similarity search for call centers to find locations of interest in transcribed calls based on semantics. Users could write in sentences or phrases and find all similar locations.
- Assisted the business in developing a suite of additional tracking metrics for ML so that model performance and business performance and attribution could be measured, including stakeholder management.
- Integrated the ML services as APIs with the existing pipelines and the application flow with a parallel engineering team.
Oregon Health & Science University
- Developed NLP models and new data transformation pipelines on large amounts of text data (for NLP predictive modeling) to create predictive models for a rare disease (Amyloidosis).
- Developed remote data pipelines in a HIPAA setting to process large volumes of healthcare data (from ~12 different patient tables).
- Created new data processing methodologies to run on arbitrary text data (from different sources) on patients in order to provide the highest predictive power of those patients that are undiagnosed.
- Created NLP predictive models to be deployed to predict real-world patients on the transformed vector spaces of healthcare data.
Database DevOps Creator for Time-series
- Understood the client's needs on time-series and developed a problem statement. Researched all available time-series databases and recommended a solution and architecture.
- Designed a Redshift database architecture to have the fastest possible access times (<500 ms) for arbitrary time-series data. Data is basically arbitrarily scalable.
- Designed a database capable of handling minute-level real-time time-series for predictive models trading cryptocurrencies.
- Wrote all the Python code and developed a custom Python package to access the database as concurrent connections to improve throughput.
- Dockerized all applications maintaining data and deployed them to AWS Fargate. Created one-line deployments to simplify ongoing improvements.
- Created a public-facing API via AWS Fargate (auto-scaling) to allow read/write/delete access to the database configured to grant permissions to individual time-series giving API keys. This allowed the client to give limited access to others.
Senior Lead Analytics
Ensemble Health Partners
- Led a small ML team developing new analytics for a large-scale healthcare company to innovate new algorithms and products while interfacing with stakeholders in the company.
- Developed novel algorithms and computational infrastructure for predictive models applied to large hospital outsourcing companies resulting in $0.1–$1.5 million in revenue lift per month per hospital (applied to dozens of hospitals).
- Developed a novel algorithm to predict erroneous hospital inpatient visit records related to ICD-10 and diagnostic-related group coding to ensure maximal profitability. The algorithm was deployed and is functional in dozens of hospital groups.
- Developed novel algorithms applied to detect missing charges in outpatient hospital visits. Invented new algorithms related to association rules and k-nearest-neighbor to increase performance.
- Established the computational infrastructure to rapidly build and deploy predictive models to dozens of health care clients. The infrastructure is now used by dozens of engineers to rapidly deploy models to all clients.
- Developed an inpatient DRG encoder (from scratch) to augment predictive models. Required understanding of the most inner workings of inpatient hospital billing.
- Wrote patents and led several developers to advance the existing infrastructure on Azure.
Department of Defense Subcontractor
- Built and trained a lipreading deep neural network with TensorFlow to predict numbers, letters, and words that were spoken by readers.
- Located and cleaned multiple training datasets for lipreading.
- Configured and set up a GPU training pipeline on an AWS EC2 infrastructure.
- Modified deep neural network construction and data pipelines to optimize for the real-world problem versus the academic problem initially posed by the client.
- Achieved the client's desired 85% requirement for accuracy.
Head of Analytics and Solution Architect
Nestlé Institute of Health Sciences
- Led a team of six to eight software and ML developers at a brand new R&D institute to drive state-of-the-art analytics on large-scale bioinformatics data combining six data outputs from different research labs.
- Invented and developed new nutrition analytics underlying a Nestlé/Samsung partnership. Built a nutrition recommendation engine leveraged largely inside Nestlé.
- Developed and designed the core analytic infrastructure for a large-scale research institute, leading a team to implement it.
- Developed bioinformatic models for integration (genomics, proteomics, metabolomics, clinical data) for multi-million dollar clinical studies.
CTO | CEO
- Contributed to the build of neuroanatomically-based systems (brains in computers) along with the relevant infrastructure.
- Created the first issued patent on simulated intelligence and neuroanatomically based systems.
- Built full-scale brain simulations that were precursors to many of the neural net architectures used today.
- Won support for early stage IARPA investment in DC.
Senior Analytics Manager
- Designed custom models (linear and non-linear) for multiple Fortune 1000 companies; this involved rewards recommendations, medical hospital visit revenue predictions, vehicle auction models, and more.
- Invented a new type of generalized predictive adaptive non-linear model combining a Kalman filter with K-NN, and built and deployed the model to production. It requires near-zero maintenance with continuous best-in-class predictions.
- Worked as a solution architect to both understand and formulate problems as well as develop rapid prototypes on the initial data.
This may be the first scientific publication published simultaneously with an interactive web app, iPhone, and iPad app.
An example of solving a fairly difficult problem: "How does the brain work?"
Python, Python 3, SQL, C
Microservices, Data Science, Microservices Architecture, Parallel Programming, ETL
Artificial General Intelligence (AGI), Predictive Modeling, Machine Learning, Algorithms, Artificial Intelligence (AI), Analytics, Big Data, Natural Language Processing (NLP), Deep Learning, Computer Vision, Build Pipelines, Electronic Medical Records (EMR), Time Series, Mobile First, MobX-State-Tree (MST), Containers, Container Orchestration, OCR, APIs, Speech to Text, Translation, Containerization, GPT, Generative Pre-trained Transformers (GPT), Gmail, GPU Computing, HL7, Electrical Engineering, Cryptocurrency, Hugging Face, Neuroscience
React, MobX, Amazon Rekognition, NumPy, TensorFlow, Python Asyncio, PyTorch
GitHub, AWS Fargate, Amazon Elastic Container Service (Amazon ECS), PyCharm, WebStorm, Slack, Azure Machine Learning
Docker, Amazon EC2, MacOS, Linux, Amazon Web Services (AWS), Azure, Kubernetes
Data Pipelines, Redshift
PhD in Integrative Neuroscience
UCSD | University of California, San Diego - San Diego, CA, USA
Master's Degree in Control Theory
UCSD | University of California, San Diego - San Diego, CA, USA
Dual Bachelor's Degrees (BSc/BA) in Electrical Engineering
University of San Diego - San Diego, CA, USA