The UK Telecoms Innovation Network (UKTIN) is working on an advanced AI toolkit aimed at helping researchers, investors, and companies navigate telecom research projects.
This innovative tool is being developed by the Smart Internet Lab at the University of Bristol and introduces a new approach to searching telecom research data.
Also Read: Amazon’s GBP 8 Billion UK Investment: What This Means for Britain’s Future
LLM x RAG
The prototype of the AI Discovery Toolkit combines a generic large language model (LLM), Google Gemini, with retrieval-augmented generation (RAG) technology. It allows users to explore a ‘data lake’ of research projects sourced from open databases, including those from UK and EU research agencies.
The toolkit’s user interface features an interactive chatbot, supplemented by transparent access to the underlying data, complete with references for better insight.
In 2020, concerns regarding telecom infrastructure security, especially related to high-risk vendors like Huawei, sparked discussions around supply chain diversification, according to Prof. Dimitra Simeonidou, UKTIN’s Lead for UK Research Capability. She noted that the telecom sector is highly fragmented, with involvement from universities, large operators, and startups but lacking major vendors. To address this, initiatives like OpenRAN and Open Networking aimed to create a stronger market for commercializing UK innovations.
Read More: BT Group and AWS Launch GenAI Gateway for AI Integration
A Central Platform for Navigation
Simeonidou emphasized the importance of a central platform to make it easier for companies to navigate the UK’s telecom landscape, and the new AI-powered Discovery Toolkit serves this purpose. The tool leverages AI and large language models to enhance data discoverability, enabling users to engage with, and transform, raw information into actionable insights.
The design of UKTIN’s AI Discovery Toolkit has advanced to incorporate large language models (LLMs) such as GPT, Gemini, and Llama. Rather than training these models on specific data, they are employed for natural language processing and paired with semantic structures known as embeddings to facilitate more intuitive communication.
By using a data-lake approach, the toolkit ensures that the information is always current, mitigating the risk of LLMs generating inaccurate or invented data, a common issue known as hallucination. Additionally, users can export the search results as a CSV file for further analysis.
The UKTIN AI Discovery Toolkit leverages the cloud-based Google Gemini LLM, selected for its capacity to handle the highest number of tokens. This allows users to access details from up to 250 projects using the 2000-token input limit. The system is hosted by Google.
The toolkit’s flexible architecture also supports the use of other LLMs, allowing users to perform less detailed queries or verify results by cross-checking with another model.
Looking ahead, the team aims to expand the tool’s capabilities to search patent text from the UK, EU, and potentially the U.S. They are also considering broadening its application to areas like semiconductor research. The prototype is available at discover.uktin.net, though UKTIN registration is required for access.