How I build · Experience · Proof of work · Wall of Mastery · Recognition · Stack · Education

I'm an AI/ML engineer at Accenture who builds production AI that shows its evidence or says it doesn't know. At work, that is a multi-agent GenAI platform running incident, problem, release and service-request operations. In public, it is a set of builds where models cite, verify and fail closed. Every claim on this page links to something you can check.

Six rules that run through everything I ship, each with a repository you can audit.

Built an enterprise AI operational assistant: a production multi-agent GenAI platform that took SME-heavy production support, incident, testing and release work and handed it to specialist AI agents.

• Multi-agent orchestration. Designed LangGraph workflows with incident, problem, release and service-request agents, routed dynamically by user intent and semantic retrieval.
• Enterprise RAG. Built indexing and retrieval pipelines over Jira, Confluence, SharePoint, GitHub, Splunk logs and databases, covering tickets, logs, documents and SQL.
• Real-time root-cause analysis. Correlates logs with historical incidents and writes context-grounded resolution steps: ≈20% faster turnaround and ≈50% less SME dependency.
• AI test generation. Turns design documents and existing frameworks into functional test cases and Selenium, UFT and Karate scripts.
• Release intelligence. Reads GitHub commits, Confluence and Jira to draft release notes, deployment plans and rollback strategies.
• Operations automation. Chat-to-ServiceNow ticket creation, docs-to-SQL analysis that pinpoints impacted tables, anomaly detection and audit-ready SOX reporting, all behind one context-aware chat backed by MongoDB.

_{Stack: Python · LangGraph · LangChain · RAG · vector search · MongoDB · PostgreSQL · Databricks · Jira · ServiceNow · Confluence · SharePoint · Splunk · Selenium · UFT · Karate}

• Built and deployed Spring Boot / Core Java microservices with scalable REST APIs for insurance workflows.
• Designed event-driven serverless components on AWS Lambda, cutting infrastructure overhead.
• Automated build, test and deploy with CI/CD pipelines on AWS CodeBuild, removing manual release effort.
• Raised reliability with JUnit unit and integration tests, and caught vulnerabilities early with SonarQube and SourceClear gates.
• Delivered in Agile Scrum and wrote the automation scripts behind a DevOps-first team.

18 public builds. The newest share one idea: models propose; deterministic code and people decide. Test counts are read from each repository; the Proof Atlas suite was run and passed for its release.

NexusRAG source · Proof Atlas live · Proof Atlas source

_{PantryProof also runs at pantryproof.vercel.app.}

74 credentials · 7 ecosystems · 5 expert · 14 professional · all issuer-verifiable

The table is ordered like the real one: light elements are fundamentals, heavy ones are expert-level. Element 1 is AZ-900. Elements 70 to 74 are five Microsoft Expert certifications, and the heaviest, AZ-305, sits in tungsten's slot. Tap any signature credential below to verify it now, or open its certificate in the vault.

_{The three Google Cloud Technical Expert badges (Customer Engagement Suite with Google AI, Intelligent Search, Build with Vertex AI) came with Early Adopter recognition: read the post.}

_{Cloud certifications cover AWS, Google Cloud and Azure architecture in depth. See the Wall of Mastery.}

B.Tech, Computer Science & Engineering: Maulana Abul Kalam Azad University of Technology, West Bengal · 2018–2022 · GPA 9.14 / 10

_{Data Structures & Algorithms · Operating Systems · Database Management Systems · Computer Networks · Machine Learning · Software Engineering · Distributed Systems}

Chess · cricket · BGMI 5187468970 · Clash of Clans #92J9PUUYU · practical problem-solving. Challenges welcome.

anupam020202@gmail.com · anupam-roy.vercel.app · linkedin.com/in/anupam--roy

_{Repository facts reflect each project as reviewed in October 2026. Visuals are generated from one data file.}