[DMP 2026]: AMRIT Agentic AI Coding Framework

[drtechie]

Description

Build an agentic AI framework for the AMRIT open-source EHR platform that embeds AI assistance into every phase of AMRIT's Software Development Lifecycle. From requirements analysis through development, code review, testing, deployment, and support, AI agents should have deep contextual knowledge of AMRIT's architecture, conventions, and project documentation so they can assist meaningfully at each stage.

The framework should work across AI coding tools (Claude Code, Cursor, Copilot, Gemini) and enable developers, QA, DevOps, and open-source contributors to get AMRIT-aware assistance without manually pasting context into every session.

AMRIT ecosystem: 10+ repositories | Angular 4-19 frontends | Java/Spring Boot backends | Kotlin/Android mobile | Confluence docs | JIRA project management | Open-source contributors via C4GT

Goals & Mid-Point Milestone

• Design the framework architecture covering MCP servers, agents, skills, commands, and coding standards
• Build or integrate MCP connectors for AMRIT's key knowledge sources (project docs, Confluence, JIRA, codebase)
• Define initial AMRIT coding standards distributable to any AI agent
• Identify and implement AI agent touchpoints across AMRIT's SDLC phases
• Mid-point milestone: At least one MCP server operational with indexed AMRIT docs, one working skill covering one SDLC phase, and a documented contribution guide
• Create a starter set of reusable skills and commands covering additional SDLC phases
• Document the framework so open-source contributors can extend it with new servers, skills, and standards

Setup/Installation

• AMRIT Setup Guide
• AMRIT Software Development Lifecycle
• AMRIT repositories: https://github.com/PSMRI

Expected Outcome

An open-source, modular agentic framework that gives AI agents deep understanding of AMRIT's codebase, documentation, and processes, and plugs AI assistance into every phase of AMRIT's SDLC. Anyone who sets up this framework should be able to ask questions about AMRIT in plain English and get accurate, context-aware answers. The framework should be extensible by design so the community can grow it over time.

Codebase and knowledge use cases:

• Ask "How does the 104 helpline call routing work?" or "What API does beneficiary registration call?" and get answers from indexed AMRIT docs and code
• New contributors run a setup command and their AI agent understands AMRIT's architecture, repos, and how they connect
• Agent understands cross-repo relationships between frontend, backend, and mobile projects via shared APIs

SDLC use cases (not exhaustive):

• Requirements/Analysis: Generate structured JIRA tickets from Confluence concept notes or BRDs
• Sprint Planning: Assist with estimation and backlog grooming using codebase context
• Development: Write code that follows AMRIT's repo-specific standards (Angular, Spring Boot, Kotlin) automatically
• Code Review: Review PRs against AMRIT's coding guidelines, flag issues across Angular, Java, and Kotlin repos
• QA/Testing: Generate test cases, assist with defect analysis, support automation efforts
• DevOps/Deployment: Assist with CI/CD workflows and deployment validation
• L2 Support: Help triage escalated bugs, search knowledge base, identify root causes
• Security/Compliance: Flag vulnerabilities, check against AMRIT's secure coding practices
• Documentation: Generate release notes, maintain knowledge base, onboard new contributors

Contributors are encouraged to propose where AI agents can add the most value and how.

Acceptance Criteria

• At least one MCP server indexes AMRIT docs and serves plain-English search queries
• At least one integration with an external tool (Confluence, JIRA, or similar)
• Initial set of AMRIT coding standards defined and distributable
• At least one working skill demonstrating AI assistance in an SDLC phase
• Framework works with at least 2 AI agents (e.g., Claude Code + Cursor)
• Documentation sufficient for a new contributor to install the framework and start using it
• Framework is modular so contributors can add new components without modifying core

Implementation Details

The framework should consist of the following building blocks. The specific implementations are open for contributor proposals.

MCP Servers connect AI agents to AMRIT's knowledge sources. Examples:

• A project docs server that indexes architecture docs, API specs, and wiki content for search
• Connectors to Confluence (concept notes, BRDs) and JIRA (tickets, sprint boards)
• A code context server for cross-repo search and API endpoint discovery

Coding Standards codify AMRIT's conventions so agents follow them automatically. Examples:

• Spring Boot service patterns (repository/service/controller layering)
• Angular component conventions (standalone components, HTTP interceptors)
• Kotlin/Android architecture (MVVM, Room, WorkManager, Hilt)

Skills are reusable agent capabilities. Examples:

• Generate JIRA tickets from a Confluence concept note
• Produce an implementation plan from a JIRA ticket
• Answer codebase questions by searching indexed docs

Commands are developer workflows. Examples:

• Bootstrap a local dev environment for any AMRIT repo
• Onboard a new repo into the framework

Open for Contribution

This ticket intentionally defines the framework's shape, not its full implementation. Open-source contributors are encouraged to propose and build:

• AI agent integrations for any SDLC phase (requirements, development, review, QA, DevOps, support)
• New MCP server integrations (Slack, GitHub Actions, monitoring, SonarQube, etc.)
• Additional coding standards for specific AMRIT modules or languages
• Domain-specific skills (clinical workflow analysis, regulatory compliance, etc.)
• IDE-specific plugins or extensions
• Novel use cases we have not thought of yet

Mockups/Wireframes

N/A. This is a developer tooling and infrastructure project.

Product Name

AMRIT

Organisation Name

Piramal Swasthya Management and Research Institute

Domain

Healthcare

Tech Skills Needed

• TypeScript
• Node.js
• Python
• RESTful APIs
• Docker
• Java
• Angular
• AI
• DevOps
• Other

Mentor(s)

@drtechie

Category

• AI
• DevOps
• Configuration
• Documentation
• Other

[drtechie]

This is a Dedicated Mentoring Program (DMP) ticket under C4GT. Applications will be accepted through proposals via C4GT, and assignees will be added through due process. Please refrain from spamming.

[zeba-source]

Hi @drtechie,

This problem caught my attention — especially the idea of making AI agents context-aware across AMRIT’s codebase, JIRA, and Confluence. I’m interested in exploring MCP-based context retrieval, reusable skills (like JIRA → implementation plans), and codebase indexing for better cross-repo understanding, along with enforcing coding standards.

I’ll be submitting a proposal via C4GT. Please let me know if there are any specific expectations I should keep in mind while drafting it.

Thanks!

[Shubhamx404]

hii @drtechie

Thanks for sharing this opportunity , this problem space genuinely interests me, especially the idea of making AI agents truly context-aware for a complex system like AMRIT. I’d really like to contribute here.

I’m thinking of a modular setup with MCP servers for docs, codebase, Confluence, and JIRA, along with reusable skills like JIRA → implementation plan and Confluence → ticket generation, usable across agents like Claude Code and Cursor.

For the mid-point milestone, I’ll aim to have one MCP server (docs/code indexed), one working skill, initial coding standards, and a simple setup flow for contributors.

Quick questions:
• Any preferred stack or prior MCP work to align with?
• Will sandbox access be available for JIRA/Confluence?
• Any priority repo/module to start with?

[drtechie]

I have edited the ticket to include more details and links to our SDLC process. This ticket is a concept and an open idea. Happy to review proposals that demonstrate genuine effort and thoughtful analysis. The less AI-assisted your proposal is, the better. Or even if AI-assisted, you should clearly understand and be able to defend what you are proposing rather than taking AI output at face value.