Infobip
About the organisation
Infobip is a global communications platform-as-a-service, founded in 2006 by three engineers who studied at the Faculty of Electrical Engineering in Zagreb and started out of a garage in Istria. The company has grown into an organisation of around 3,500 employees across 75 offices, 40-plus data centres and over 800 direct operator connections, handling around 42 billion monthly interactions across SMS, voice, WhatsApp, RCS and chat applications on behalf of clients including Uber, T-Mobile and UNICEF. Above that infrastructure sits a portfolio of products including Infobip Answers, the company’s low-code, omnichannel chatbot building platform. A dedicated AI Hub and a research and development arm work on conversational experience, human-AI interaction, network optimisation and fraud detection. Infobip is also Croatia’s representative in the IPCEI Next Generation Cloud Infrastructure Services consortium, helping to build secure, energy-efficient and privacy-first European communications infrastructure. Innovation sits at the structural core of how the company operates: the first GRAPHIA / LUMEN Innovation Prototyping Lab (IPL) was hosted at Infobip’s Zagreb campus precisely because the founders designed the campus as a venue for cross-organisation exchange.
What challenges can the GRAPHIA project help solve?
Infobip’s research and engineering teams build conversational agents that need to ground their answers in trustworthy, structured knowledge. The same pattern recurred throughout the IPL: clients arrive with documents, policies and product information and expect a chatbot that can answer questions accurately and traceably, often in regulated domains where errors carry real cost.
The team’s own experiments include constructing a knowledge graph from a code base, using LightRAG, to give coding assistants the structural grounding that pure vector retrieval cannot provide. GRAPHIA matters to a communications platform because a federated SSH knowledge graph offers exactly the kind of high-quality, ontology-anchored, provenance-bearing source that current retrieval-augmented generation systems lack when applied to social and humanities questions.
GRAPHIA represents a curated, pluralistic, peer-grounded source of SSH evidence at a scale rare in the commercial knowledge-base ecosystem. For Infobip’s research team, the project is also a reference point for large-scale, federated, semantically interoperable knowledge built to standards of provenance and verifiability that industrial users increasingly expect.
What is the proposed use case?
The Day 3 sessions at the IPL in Zagreb included open discussion of what a conversational interface over the GRAPHIA SSH knowledge graph might look like, delivered through the kind of platform Infobip already operates at industrial scale. In an arrangement of that sort, researchers, students, journalists and policy users could imagine interacting with the federated SSH corpus through the channels they already use, including web chat and messaging applications, with their questions translated into structured queries against the graph and the evidence returned in plain language with citations and confidence indicators. Several of the technologies Infobip presented at the IPL would be candidates for any future prototype. The proprietary chunker shown by the team, evaluated on legal, pharmaceutical, telecoms and retail documents at over ninety-five per cent success and offered free for experimentation, could in principle be applied to SSH ingestion. The agentic context-engineering pattern presented in the Day 3 workshop, which reduces multi-turn reasoning to a sequence of single-turn state updates, maps onto the multi-step nature of research questions.
What type of SSH data or content are involved?
Infobip is, in this use case, a content consumer and conversational delivery layer. The SSH content of interest covers the breadth of disciplines that the GRAPHIA knowledge graph will federate: sociology, cultural studies, social psychology, history, media studies, applied ethics, anthropology, and policy-adjacent humanities, alongside cross-disciplinary materials linking SSH to industrial questions such as communications behaviour, generational shifts in messaging, and human factors in AI adoption. Infobip’s research team has direct interest in human-AI interaction studies, conversational experience research, and engagement behaviour, all of which sit within or adjacent to SSH. Beyond surfacing GRAPHIA content for end users through the company’s communication channels, there is potential for some of Infobip’s own published research outputs (for example the team’s recent work on gamified multimodal chatbots) to be considered as inputs to the SSH evidence base where appropriate.
What technical aspect would be used?
Both content and technical dimensions are involved. On the technical side, the IPL sessions revealed that three GRAPHIA capabilities are relevant for Infobip’s current work. First, the federated SSH knowledge graph itself, as a grounding layer for conversational AI applications that handle questions about people, society, culture and behaviour.
Second, the AI modules being built around the graph by GRAPHIA partners, including the text-to-SPARQL agent demonstrated by Odoma, with its disciplined approach to context engineering through VoID descriptions, ShEx shapes, and representative examples. Third, the broader hybrid AI architecture pattern of combining structured knowledge with large language models, which the Day 2 panel and the Day 3 agentic systems workshop both addressed at length.
Technologies most readily applicable from Infobip’s own portfolio include the Infobip Answers chatbot platform as an interface tier, the document chunker as a possible enrichment tool for SSH ingestion, and the agentic ReAct-pattern workflows demonstrated for the Dungeons and Dragons engine and the car-parts assistant, and the in-house agentic e-commerce conversational recommender and shopping assistant, which is relevant for guided discovery, intent capture and recommendation over structured catalogues. Engagement at this stage is exploratory and further conversations are needed to determine which combinations are worth prototyping.