History
Protofect was born in New York City, as a way to address the various challenges that data science and artificial intelligence solutions face when exposed to real world. Ranging in issues from implementations, deployments, instrumentation, ingestion, modeling and tuning to quantifying the ethical implications of real world artificial intelligence models on its users - we stand on the bleeding edge of whats possible with AI.
The team possesses extensive expertise in the hidden nuances of building ML solutions, such as understanding the brittleness of models, the required tuning efforts for hyper-parameters and design for AI, impact and scale of AI products.
Vision
Learn more about our approach through our blog.
measuring ai performance ▸
Principles
The various segments Protofect touches
with great power…
Beyond the data and the model
Machine learning focuses on computational structures and statistics. And data science works similar to modeling the airflow surrounding an airplane (here the airplane is your product). But AI solutions need to address a wider spectrum of things beyond the model, the product or even the company. Depending on impact of your company, there are several considerations - from security and explainability to lineage. Our hearts and minds are aware of the broader responsibility.
Mission
Your solution could impact the life of one person deeply or thousands superficially. They might be aware or not, of an AI’s presence when interacting with software.
AI engines have a deep responsibility in protecting the welfare of its consumers.
Building
• Data Lineage, Adversarial Security
• Supervised, Unsupervised and Transfer Learning
• Explainability and Data Stories
Advising
• Collaborative AI, Market Impact, Alignment
At every phase…
Redefining the intelligence pipeline
AI feeds on data; good data. Data must not only be collected with powerful sensors and estimation algorithms, but also protected so models don’t get compromised. We have extensive experience at every stage of the pipeline, whether its instrumentation for proper data collection, or time-series models that need constant updating, or interactive visualization which stakeholders can play with while accommodating for failsafes, rollbacks, unit tests (specific to ML software) or lifelong learning.
Critical Research
Patents/ Papers
Automatic Conversational Scoring
Latent Emotion Understanding
Exclusive/Adversarial test datasets
AI level Ratings
Activities & Affiliations
• Prescriptive instrumentation in data collection
• Modular approach to AI pipelines
Special Focus
• Lifelong learning, Bias identification, Interactive AI
roy@protofect.com
who we are…
Leadership
SD Roy spent several years working with computational structures and understanding how real world systems can be modeled and improved with artificial intelligence. His interest ranges a spectrum of industries & markets with a common thread - they all are moving towards data driven and AI solutions and need predictive augmentation at massive scale with minimal human intervention in the loop.
With a PhD in AI/ML that modeled transfer learning for social-semantic web, and strong experiences at Microsoft Research, Betaworks, Columbia Journalism School etc. He is joined by Jeff, former Navy engineer and Claudio, a full stack developer.