Rajkumar Sakthivel is a senior backend engineer at Tesco and co-creator of Code Context Engine, an open-source system that reduces the amount of source code AI coding assistants send to language models. His focus is the expensive, often overlooked input side of AI-assisted development: irrelevant context increases costs and can make generated answers less reliable.
Based in London, Sakthivel has worked across PHP, Laravel, Python, observability, cloud infrastructure, and enterprise backend systems. He studied at the University of Oxford between 2022 and 2024 and builds production AI systems involving sentence-transformer models, language-model integrations, and private-cloud infrastructure. His public technical work developed from simplifying DevOps tooling and processes in 2024 to agentic AI systems in 2025 and developer-tool infrastructure in 2026.
How he approaches AI coding infrastructure
- Fix excessive input context. Changing prompts, output limits, or model settings cannot recover the cost of unnecessary source code already supplied to a model; relevant context must be selected before inference.
- Combine structural and hybrid retrieval. Code Context Engine uses local code indexing, Tree-sitter parsing, hybrid semantic and keyword retrieval, and cross-file relationships to return relevant functions, classes, and methods instead of complete files.
- Make relevance filtering cheap. Lightweight scoring combines semantic similarity, keyword matches, and code recency without requiring another language-model call.
- Share project knowledge across assistants. A common index and cross-tool project memory let Claude Code, Cursor, Copilot, and similar tools reuse repository context through the Model Context Protocol.
In a public FastAPI benchmark, Sakthivel reported approximately 94% fewer input tokens across 53 files and 20 developer questions. That figure compares against repeatedly reading entire files, not modern coding agents’ smarter retrieval; he also acknowledges that performance can deteriorate substantially on larger, less cleanly organized codebases.
