Tech Fabric is a Digital Transformation firm that has been serving to enterprises embrace the most recent developments in expertise to enhance their operational efficiencies, scale back prices, and ship worth to their inner workers or prospects.
Our prospects depend on our steering relating to the expertise panorama and the way they will leverage newer technological developments to be more practical, disruptive, and aggressive.
Whether or not it’s implementing higher methods of managing information, offering higher options to combine with their companions, or constructing actually nice UX for his or her methods, we’ve all the time strived to be their trusted adviser and convey them worth by way of the introduction of the suitable structure, instruments, frameworks, or options to handle their wants and assist them keep forward of their competitors.
With the current developments in Generative AI house, we’ve had many conversations with our shoppers on how they will make the most of Giant Language Fashions and introduce Generative AI experiences of their purposes that may democratize entry to information, simplify their present workflows, and assist their workers be extra productive.
Since each consumer is completely different and their wants are completely different, we regularly needed to construct {custom} options tailor-made for his or her wants and located ourselves doing the identical factor repeatedly. Since there are various parts and transferring items in a Generative AI system, there’s no option to construct a one-size suits all system. We select the parts based mostly on price, pace, and options which can be applicable for our shoppers.
A typical Generative AI system can have these constructing blocks and relying on the corporate and their use case, we regularly have to decide on completely different parts, typically spreading throughout native cloud supplier parts to third-party and custom-built parts.
- Vector Database (Azure AI Search, Quadrant, Cloudflare Vector Retailer, Elastic Search, Vertex AI Search and many others.,)
- Doc Parsing Engine (Azure Doc Intelligence, Llama Parse and many others.,)
- AI Mannequin Serving Platform (Azure OpenAI, AWS Bedrock and many others.,)
- Data Graphs
- Clever Information Platform (Databricks, Snowflake, Azure Material, and many others.,)
- File Storage (Azure Blobs, AWS S3)
- Retrieval Augmented Technology (RAG) methods
- Workflow Orchestration for steady coaching (Temporal)
- Id Entry Administration (Azure Entra ID, Ping, Okta and many others.,)
and plenty of extra…
As you may see, there are a number of core parts which can be completely wanted for each significant Generative AI platform. However not each enterprise makes use of the identical cloud platform, has the identical set of necessities and even tries to resolve the identical downside.
A few of them have tons of paperwork in unstructured format (PDFs, spreadsheets, phrase docs, handwritten notes) accrued over many years. A few of them have clever information platforms with superior capabilities. A few of them have semi-structured information and conventional relational databases.
Every situation requires utilizing a distinct type of parsing engine that’s extra suited to our prospects’ wants, a distinct vector database, a distinct foundational AI mannequin that’s fantastic tuned for his or her use case, and many others.,
For that reason, we regularly discover ourselves attempting to construct a {custom} platform for every of our shoppers with guide setup and integration. This strategy isn’t future proof, isn’t price efficient for our prospects, and may shortly go old-fashioned, contemplating the tempo at which innovation is occurring on this house.
There’s bought to be a greater option to quickly provision infrastructure and swap out parts as newer, higher parts arrive. We additionally need to have a platform that may host a number of tenants and gives an incredible person expertise (UX) to ingest, course of, prepare and deploy AI chatbots and copilots on-the-fly.
That’s why we determined to construct Fiber, a brand new multi-tenant, multi-cloud, and multi-model AIOps platform, that can be utilized to shortly create and deploy AI chatbots and copilots skilled on enterprise information.
Customers can have full flexibility to decide on numerous choices for parts within the system and provision underlying infrastructure on the click on of a button, decreasing months’ value of labor to few hours.
For instance, they’ll have an possibility to decide on Elastic Seek for vector search, Azure Doc Intelligence for parsing engine and Azure OpenAI for mannequin serving and deploy it to Azure. Or they may select Qdrant for vector search, LlamaParse for parsing engine and AWS Bedrock for mannequin serving and deploy it to AWS. They’ll have many extra choices to select from native cloud suppliers choices to 3rd social gathering and opensource choices.
It’s their selection. What platform they need to deploy to and what parts greatest match their use case is for as much as them to decide on. Fiber gives them complete management over their information privateness and safe entry to cloud infrastructure.
Fiber will allow the businesses to shortly create chatbots and copilots skilled on their proprietary information and deploy it throughout their Enterprise by way of granularly managed permissions and safety insurance policies.
Right here’s excessive stage structure.
We’ve determined to construct a multi-tenant system that may be simply provisioned whereas totally preserving our prospects’ information privateness and management over their information.
Every tenant can have all their assets provisioned in their very own cloud subscription. All their information, vector shops, AI fashions, parsing engines shall be provisioned of their cloud platform of selection. Fiber purposes and APIs shall be granted granular entry to assets based mostly on the permissions granted within the Id and Entry Administration System authenticated with OAuth 2.0. Your complete infrastructure provisioning course of is automated and orchestrated with Temporal workflows and its sturdy execution capabilities.
Since we have now so many dependencies on exterior cloud suppliers and their APIs, we would have liked to have a sturdy and deterministic execution of processes in our system. Temporal sturdy execution framework gives a assure that it doesn’t matter what occurs — course of crashes, community or storage outages, its orchestrator helps the failed processes get better by rehydrating them in a distinct server and proceed the circulate of execution.
Temporal framework is the spine for all our underlying workflows. We execute workflows throughout onboarding new shoppers, and customers, provisioning infrastructure, coaching AI fashions, creating vector embeddings and many others.,
The great thing about Temporal framework is that it really works with our pre-existing decisions for runtime, check framework, CI/CD environments, and any net framework. We don’t want to decide on a particular language or server expertise to utilize Temporal workflows. We now have workloads written in Python, Typescript, C#, .NET Core, Bicep DSL, Terraform, and many others., and Temporal simply works with each a type of applied sciences.
There’s no different workflow orchestration engine that’s as versatile as Temporal and helps all the key programming languages. It’s really a pleasure to work with it and we couldn’t be extra excited.
This versatile structure of Fiber permits us to shortly provision Generative AI platform for our shoppers to allow them to leap proper into creating AI chatbots however can be future proof. If newer fashions arrive, or there’s a greater vector search database, or newer parsing engine works higher, it’s solely a matter of some clicks to swap these parts and Temporal workflow will do the remainder to retrain the AI mannequin and populate the vector retailer!
Try Fiber Copilot and reach out for those who’d wish to attempt it out!
—
Preetham Reddy
Founder — Tech Fabric and Fiber
