T2R2
Reproducibility Crisis
It might seem surprising as it’s all about computers and mathematics, but it is a well documented fact.
It’s a substantial part of the current general reproducibility crisis in Science.
AI and Data Analysis are cumbersome. Producing a Yes or No decision for a drug, a vaccine or a device for one patient or thousands implies running millions lines of code written in many different programming languages.
The smallest change in a single step can have a huge impact making a result not Reproducible.
T2R2 solves the problem at its core. It bridges the gaps between Research and Development enabling true Translational AI and Data Science.
Decentralized, Traceable and Reproducible Platform
T2R2 stands for Transparent, Traceable, Replicable and Reproducible by design.
T2R2 is a set of software bricks that can be assembled and extended to develop decentralized production ready AI solutions.
We started T2R2 for the Life Science Industry where huge amounts of data (e.g. TBs per study in Genomic or proteomic) need to be processed every day. As reproducibility is a principle that underpins the scientific method, results need to be sound and traceable.
All available solutions so far are extremely centralized and proprietary. When there, traceability and reproducibility features have been added as afterthoughts.
Also in recent years, the cloud centralized paradigm is shifting to more decentralized and distributed architectures.
An obvious reason is the cost of transferring data from local laboratory to remote cloud infrastructure. Especially as some local machines are now more powerful than cloud nodes a couple of years ago.
As an example, T2R2 enables to run an alignment on NVIDIA GPU local machine instead of sending sequencing data to the remote server. Still, all the traceability and reproducibility features are provided.
From the very beginning, T2R2 has been designed as a decentralized and distributed solution (aka multi-agent) that provides Traceability and Reproducibility by design.
Multi-Agent Architecture
The core architecture of T2R2 is based on the Multi-Agent paradigm.
Each agent has a particular well defined role. Agents can be combined in all possible ways into complex multi-agent systems that can also become AI Graph-of-Thought systems.
T2R2 agents can implement features or connect to remote services.
They can also act as side cars of already implemented feature in any programming language.
They can be added in a container as wrapper or as "helper" providing features around traceability and observability.
Thus, T2R2 can instrument a module and ensure it's executed in the best conditions, with the right input data and observation level to ensure ultimate traceability.
From Edge to Private and Public Cloud
T2R2 is built using last generation of Cloud-Native Software Architecture.
It ensures jobs distribution with the highest performance for the lowest cost and power consumption.
It scales up horizontally as well as vertically on demand.
Jobs can run on Edge devices, private infrastructure as well as public cloud facilities.
AI pipelines can be deployed throughout the cloud continuum, with steps running on local nodes while others running on public cloud infrastructure.
Powered by AI for AI
T2R2 contains AI modules to optimize cloud resources usage for big data processing modules using historical data from previous runs.
AI deployed algorithms are optimized and workflows executions are continuously improved.
T2R2 is used to build complex chains of thoughts.
It includes agents to talk to LLMs, to cache vectors, to access vector databases or to manage knowledge graphs.
Immutable Backend Journal
T2R2 takes transparency to its deepest ground. Everything is tracked and nothing can be tampered.
It supports quality standards like GxP and 21 CFR Part 11 and it implements FAIR Principles as well.
The code execution becomes its own mathematical proof and is stored in a data store using Ledgers Technologies.
T2R2 provides a high performance backend based on the Kalix platform. It stores all events and information about every step in AI pipelines into an immutable journal.
Thus, T2R2 offers ultimate traceability of both workflows and results.
