Token Efficient Data Serialization for GenAI Applications
As organizations embed Generative AI into production workflows, token-based pricing and finite context windows make the structure of data exchange a first-order design concern rather than a low-level implementation detail. Traditional formats like JSON are verbose and not optimized for how LLM tokenizers segment text, leading to unnecessary cost and reduced effective context for structured payloads. TOON (Token-Oriented Object Notation) was introduced as a token-conscious, human-readable encoding of the JSON data model that leverages indentation and tabular layouts to cut 30–60% of tokens versus JSON while remaining easy for models to parse. However, TOON is optimized for flat and tabular structures and can, in some cases, cost more tokens than JSON compact for deeply nested data common in real-world GenAI applications.
This presentation introduces Line Based Object Notation (LBON), a new token-efficient serialization format designed for LLM-centric systems that provides optimized representation of both hierarchical and tabular payloads. Compared with TOON, the format reduces token usage by 14% on nested structures and 7% on tabular data while preserving deterministic lossless roundtripping. This talk will outline the key design ideas and evaluation methods, share benchmark results, and provides practical guidance for integration with real-world GenAI applications.
Dr. Srujan Kotikela is a Clinical Assistant Professor in the Department of Information and Operations Management at Mays Business School, Texas A&M University, where he teaches courses in network infrastructure, cloud computing, and cybersecurity management. His research interests span artificial intelligence, blockchain technology, cybersecurity, decentralized identity systems, and knowledge graphs.
His recent work focuses on practical applications of AI, including knowledge graph‑enhanced retrieval‑augmented generation (GraphRAG), prompt engineering, and AI‑powered phishing detection, with an emphasis on making GenAI systems robust and cost‑efficient. He has published research on cloud security vulnerabilities, decentralized identity protocols, the impact of prompt tone on LLM accuracy, and now explores token‑efficient data representations such as Line Based Object Notation (LBON) for large‑scale GenAI deployments.
We use cookies to analyze website traffic and optimize your website experience. By accepting our use of cookies, your data will be aggregated with all other user data.