Opsium
Client-side log analysis & RAG platform
Key Achievements
Architected a client-side Retrieval-Augmented Generation (RAG) pipeline utilizing a vectorless database engine to query operational logs without database hosting or embedding model overhead
Designed an in-browser BM25 and TF-IDF search engine to index, rank, and inject relevant log chunks directly into LLM prompts based on metadata weights and length penalties
Orchestrated multi-step LLM prompt chains using LangChain to automatically synthesize incident timelines, perform root-cause analysis, and generate preventative action-item checklists
The Challenge
Traditional log analysis platforms require heavy database indexing, dedicated infrastructure, and expensive embedding models, making postmortem logs expensive to query and maintain. For quick, ad-hoc incident analysis, developers need a lightweight, serverless tool that respects privacy and handles megabytes of log files entirely in the browser.
Strategy & Implementation
By implementing the entire RAG pipeline in-browser, we eliminated server hosting fees and database setup steps. The search index relies on standard, mathematical keyword relevance algorithms (BM25 and TF-IDF), which perform exceptionally well on structured operational logs. LangChain manages the prompt chains to invoke Gemini API models, providing a reliable and lightning-fast analysis framework entirely client-side.
Impact & Results
Enables instant incident timelines and postmortem report generation directly from raw log files uploaded by developers. Yields zero server database overhead and zero embedding compute latency, reducing operational log auditing costs by 100%.