Ayush Ranjan Roy

Ayush Ranjan Roy

AI Engineer at Accenture. Multi-agent systems, MCP, and human-in-the-loop workflows.

I build multi-agent systems for enterprise finance workflows (Order-to-Cash / Dispute & Deduction) with LangGraph, MCP tool access, and human review. Previously platform engineering on Accenture’s MxDR / Adaptive MxDR product on AWS — including OpenTelemetry, Datadog, and SIEM telemetry.

Claude Certified Architect — Foundations · Anthropic, May 2026

Experience

  1. Aug 2025 — Present

    AI Engineer · Accenture

    GenAI & Data

    • Multi-agent systems for enterprise Order-to-Cash / Dispute & Deduction using LangGraph, FastAPI, and PostgreSQL — durable state so long-running workflows resume reliably.
    • MCP tool access so agents call production tools under policy instead of unconstrained side effects.
    • Human-in-the-loop gates so uncertain or high-impact steps pause for analyst approval before the run continues.
    • Multi-model inference (Claude, GPT, Azure) with async execution, routing, retries, and latency / token / cost metrics for evaluation.
    LangGraphMCPHITLClaudeGPTAzure AIFastAPIPostgreSQL

  2. Aug 2023 — Aug 2025

    Platform Engineer · Accenture

    Security · MxDR / Adaptive MxDR

    • AWS platform for MxDR → Adaptive MxDR: Terraform, CI/CD, and production support across shared services and multi-account environments.
    • Redis-backed shared metadata services; Lambda upgrades (Python 3.8 → 3.12); multi-environment operations.
    • Secrets Manager and IAM automation; service hardening; restricted-egress proxy automation.
    • Observability on the security platform: OpenTelemetry health paths, Datadog monitoring, mTLS telemetry into enterprise SIEMs; blue-green deploys on live upgrades.
    AWSTerraformRedisOpenTelemetryDatadogSecrets ManagerIAMCI/CD

Selected work

  • Multi-agent finance workflows

    Enterprise · GenAI

    LangGraph multi-agent systems for Order-to-Cash / Dispute & Deduction — MCP tools, human review gates, FastAPI, PostgreSQL. Contributor on a multi-engineer team.

    LangGraphMCPHITLFastAPIPostgreSQL

  • MCP + human-in-the-loop

    Enterprise · GenAI

    MCP servers and tool bindings for governed agent tool use, with human approval before high-impact actions.

    MCPHITLLangGraph

  • Multi-model inference

    Enterprise · GenAI

    Async inference across Claude, GPT, and Azure with concurrency control, retries, routing, and latency / token / cost metrics.

    ClaudeGPTAzure AIAsync Python

  • MxDR → Adaptive MxDR on AWS

    Enterprise · Platform

    Platform modernization: Terraform, shared services, CI/CD, secrets, plus OpenTelemetry, Datadog, and SIEM health telemetry.

    AWSTerraformOpenTelemetryDatadogRedis

  • HelixOps

    Open source

    Incident IDE: multi-agent RCA over runbooks, then human-approved fixes. Mock-safe by default.

    LangGraphRAGFastAPIHITL
    Source

  • RAGGym

    Open source

    Retrieval practice platform: swappable models and stores, self-correction, evaluation harness.

    RAGLangGraphStreamlit
    DemoSource

About

AI engineer at Accenture (GenAI & Data). I build multi-agent systems with LangGraph, MCP for governed tool calls, human-in-the-loop review, FastAPI, and PostgreSQL — on multi-engineer teams shipping enterprise finance workflows.

I also build multi-model inference (Claude, GPT, Azure): async execution, routing, retries, and latency / token / cost metrics.

2023–2025: platform engineer on Accenture Security’s MxDR → Adaptive MxDR product — Terraform, shared services, CI/CD, secrets, and production observability (OpenTelemetry, Datadog, SIEM) on AWS.

Open source: HelixOps, RAGGym. Gurugram. Open to senior AI roles.

Education

B.Tech, Electronics and Communication Engineering · VIT Chennai, 2019–2023

Artificial IntelligenceEssentials of Machine LearningCryptography and Network SecurityInformation Security Analysis and AuditData Structures and AlgorithmsOperating Systems

Agents & GenAI

Multi-agent systemsLangGraphMCPHuman-in-the-loopRAGClaudeGPTAzure AIFastAPIPostgreSQL

Platform & observability

AWSTerraformIAMSecrets ManagerCI/CDOpenTelemetryDatadogDockerRedis

Contact

Best reached by email. Open to senior AI roles.