Mukesh Rajput

Problem

CI/CD automation failures consumed a large share of senior QA time: every failure required manually reading logs, classifying the type, determining a fix, and updating tests. At scale, this was unsustainable.

Solution

Built an autonomous QA AI Agent using OpenAI GPT-4 and LangChain. The agent reads CI failure logs → classifies the failure → if automation-related, generates a fix → validates it → raises a PR automatically. This first version has since been rebuilt on Claude as the Test Triaging and Test Healing agents in the QA Agent Network.

Results

Before vs After

Problem

The existing automation suite at Aspire took 10 hours for a full run, blocking CI/CD feedback loops and forcing engineers to wait overnight for results.

Solution

Replaced the framework entirely. Redesigned test architecture for parallelism, eliminated flaky tests, introduced smart priority-based grouping, and integrated per-PR feedback directly into CI/CD.

Results

Before vs After

Problem

Gojek acquired four companies: Midtrans (online payment gateway), Kartuku (offline POS payments), NadiPos (POS devices), and Mapan, each with their own QA teams, frameworks, tooling, and culture. Four separate realities that needed to become one.

Solution

Led the end-to-end merger of all four QA orgs into a single Centralised QA Function. Audited every team's processes, identified overlaps, standardised frameworks and tooling, unified reporting, and ran cross-company upskilling programmes to bring manual testers up to automation standard.

Results

• Single unified QA strategy across 4 acquired companies and 10+ product verticals
• 120+ QAs aligned under one framework, one process, one reporting structure
• Eliminated redundancy in tooling and test coverage across the merged org
• Significantly raised technical depth: manual testers from acquired companies trained in automation

Problem

Company knowledge was fragmented across Confluence, Google Drive, and Slack. Engineers wasted hours searching for documentation, runbooks, and processes.

Solution

Built an in-house ChatGPT-like platform using Flask, LangChain, and ChromaDB. Ingests documents from multiple sources, creates vector embeddings for hybrid retrieval. Slack bots (DevopsBot, TechSupportBot) serve as natural-language interfaces to the knowledge base.

Results

• Engineers query internal knowledge in natural language
• Supports GPT-4 (cloud) and Ollama (on-prem) for cost flexibility
• Open-sourced, now used by external teams too