Opsium – Vectorless RAG Postmortem Platform

Overview

Production incidents often generate thousands of log entries, making it difficult for developers to identify the root cause quickly. Traditional AI-powered log analysis platforms rely on vector databases and embedding models, which increase infrastructure costs and retrieval latency.

Opsium is an AI-powered postmortem platform that eliminates the need for vector databases by using a Vectorless Retrieval-Augmented Generation (RAG) pipeline. It combines BM25 and TF-IDF retrieval techniques with Gemini AI to deliver accurate root cause analysis, incident summaries, and actionable debugging recommendations.

Problem Statement

When a production issue occurs, engineers typically spend hours manually searching through large log files to identify:

Root cause of the failure
Affected services
Error patterns
Possible fixes

Existing RAG solutions require vector databases, embedding generation, and additional infrastructure, making them expensive and complex to deploy.

The Solution

Opsium performs intelligent log retrieval without embeddings.

Instead of converting logs into vectors, the platform indexes logs using BM25 and TF-IDF search algorithms directly in the browser. Relevant log chunks are retrieved based on keyword relevance and metadata before being sent to Gemini AI for contextual analysis.

The result is a lightweight, fast, and cost-effective AI assistant for production debugging.

Key Features

Vectorless Retrieval using BM25 & TF-IDF

AI-powered Root Cause Analysis

Automatic Incident Summaries

Fast Client-side Log Indexing

Context-aware Debugging Insights

AI-generated Resolution Recommendations

Lightweight Deployment (No Vector DB Required)

High Search Relevance with Lower Infrastructure Cost

Tech Stack

Frontend

React 19
TypeScript
Tailwind CSS

AI & Retrieval

LangChain
Gemini API
BM25
TF-IDF

Architecture

Vectorless RAG
Client-side Indexing
Metadata-based Retrieval

How It Works

User uploads production logs.
Logs are indexed locally using BM25 and TF-IDF.
The retrieval engine ranks the most relevant log chunks.
Selected context is passed to Gemini AI via LangChain.
The AI generates:
- Root cause analysis
- Incident summary
- Debugging insights
- Recommended fixes

Challenges Solved

Eliminated dependency on expensive vector databases.
Reduced retrieval latency by avoiding embedding generation.
Built a lightweight architecture that runs efficiently with minimal infrastructure.
Maintained high retrieval relevance using keyword-based search techniques.

Impact

Reduced infrastructure complexity by removing vector databases.
Improved retrieval speed with client-side indexing.
Enabled developers to diagnose production incidents faster using AI-assisted log analysis.
Delivered a scalable and cost-effective alternative to traditional RAG systems.

Future Enhancements

Multi-file log correlation
Real-time log streaming
Team collaboration dashboard
Kubernetes and cloud monitoring integrations
Support for multiple LLM providers (OpenAI, Claude, Llama)