Executive Summary
A confidential SaaS client in the engineering and project-management tooling space partnered with Meritorious CodeCrafters to build AeroFlow (internally referenced as AeroBoard), an AI-assisted agile team workspace that extends a traditional Kanban-based issue board into a full collaboration platform. AeroFlow combines sprint planning, active developer time-tracking, role-based access control, and a three-layer AI assistance and retrieval-augmented generation (RAG) system, letting developers query internal policies, update issues, and log time using natural language. Meritorious CodeCrafters delivered a decoupled full-stack application - a Next.js frontend and an async FastAPI backend - with a modular AI architecture grounded directly in the team's own operational database and internal documentation.
Client Overview
Confidential SaaS Client - Engineering / Project Management Tooling Sector
The client is building a next-generation agile team workspace aimed at software teams who currently rely on tools like Jira for sprint and issue management but want time-tracking, onboarding, and AI-assisted workflows unified into a single platform rather than stitched together across separate tools. Due to a confidentiality agreement, the client's name and specific brand identity are withheld in this case study.
Business Challenge
Software teams typically juggle multiple disconnected tools to manage delivery: an issue tracker for sprint planning, a separate time-tracking tool for worklogs, internal documentation scattered across wikis, and no easy way for developers to get quick answers about team policies or their own task status without leaving their workflow. The client identified these specific friction points:
- Traditional Kanban tools handle task management but not integrated time tracking
- Developers lack a fast, natural-language way to interact with their own tasks, worklogs, and team policies
- Onboarding new team members to internal processes and documentation is manual and slow
- Role-based access needs to be enforced consistently across sprint planning, invitations, and reporting
- Internal knowledge (policies, onboarding guides) is disconnected from the tools developers use daily
The client needed a single platform that unified issue tracking, time tracking, sprint planning, and role-based security with an AI layer capable of acting on real backend data and answering questions grounded in the team's own documentation - not generic AI responses.
Project Objectives
- Build a Jira-style Kanban board with drag-and-drop issue management and integrated time tracking
- Provide a backlog view with filtering, quick-add, and issue management
- Support full sprint planning with scheduling and multi-user sprint assignment
- Implement role-based access control distinguishing managers from employees
- Build a global timer system that tracks active work per user and converts sessions into persistent worklogs
- Create an onboarding and invitation system with role-scoped access
- Deliver time analytics showing logged vs. estimated hours and team activity
- Build a three-layer AI architecture combining a provider-agnostic AI service, a bounded-tool agent that can safely act on backend data, and a RAG system for grounded document search
Solution
Meritorious CodeCrafters designed and built AeroFlow as a decoupled full-stack application, pairing a modern Next.js frontend with an asynchronous FastAPI backend and a purpose-built, three-layer AI system.
Sprint Board and Kanban Experience
The core of AeroFlow's frontend is an interactive, drag-and-drop Kanban board built with native HTML5 drag-and-drop APIs, spanning the standard To Do, In Progress, Review, and Done columns. Issue cards include inline play/stop controls, letting developers start and stop time tracking directly from the board without switching context.
Backlog and Issue Management
A dedicated backlog tab gives teams a standardized view of all issues, with instant filtering by status and priority, an inline quick-add drawer for creating tasks with estimates, and dropdown controls for updating or deleting issues - reducing the number of clicks needed for everyday issue management.
Global Time Tracking
A sticky global timer header keeps the currently running issue timer visible from anywhere in the app, showing elapsed minutes with play/pause controls and a stop action that prompts the user to log notes. On the backend, timer operations support starting, pausing, and stopping trackers both globally and per user, automatically converting completed timer sessions into persistent worklog entries tied to the relevant issue.
Task Detail and Worklog Tracking
A sliding, Jira-style task detail drawer surfaces issue descriptions, assignee configuration, an interactive subtask checklist with a live progress indicator, and a full worklog timeline - giving users a complete history of work performed on any given issue without leaving the board.
Time Analytics Dashboard
Real-time KPI blocks compare actual logged hours against estimated hours, supported by comparative bar graphs, priority distribution breakdowns, and an audit feed of recent work entries - giving managers visibility into team velocity and workload distribution.
Sprint Planning and Role-Based Security
On the backend, full CRUD endpoints support issues, subtasks, worklogs, and sprints, with sprints schedulable and assignable to specific employees. JWT-based authentication distinguishes manager and employee roles, restricting sensitive endpoints - such as sprint creation and workspace invitations - to managers only.
Onboarding and Invitations
An invitation system allows managers to send role-scoped email invites to new team members. When an invitee accepts, their account automatically adopts the role the manager assigned, streamlining onboarding into new workspaces without manual role configuration after the fact.
Three-Layer AI Architecture
Meritorious CodeCrafters built a modular, three-layer AI system that grounds the assistant directly in AeroFlow's own operational data:
- Layer 1 - Core AI Service: A provider-agnostic client that routes requests to the Groq API (using the llama-3.3-70b-versatile model) for AI processing, with a deterministic, dependency-free Python-based embedding fallback for local cosine-similarity search when needed.
- Layer 2 - AI Agent with Bounded Tools: A chat agent that translates natural-language requests into real backend actions - fetching a user's personal tasks, updating card status, or generating invitations - through a constrained, safe set of tools rather than open-ended database access.
- Layer 3 - RAG Core: A document ingestion pipeline that parses uploaded documentation, splits it into semantic chunks, and generates embeddings, allowing users to query internal policies and onboarding guides and receive answers grounded strictly in the ingested document context rather than general AI knowledge.
Technology / Expertise
Frontend: Next.js 16 (App Router), TypeScript, Tailwind CSS v4, custom lightweight SVG icons
Backend: Python 3.10+, FastAPI (async ASGI), SQLAlchemy 2.0 (async queries), Pydantic v2
Database: PostgreSQL, with a SQLite-compatible fallback schema for local development
Rate Limiting: SlowAPI (token-bucket rate limiting per route)
AI / LLM: Groq API (llama-3.3-70b-versatile) for agent tool execution, with a deterministic local embedding fallback for similarity search
Data Model: SQLAlchemy models spanning users, sprints and sprint members, issues and issue labels, subtasks, work logs, active timers, invitations, documents and document chunks, and AI session/message history
Implementation Process
01. Discovery and Architecture
Meritorious CodeCrafters worked with the client to define AeroFlow's scope: a Jira-style issue board extended with native time tracking, sprint planning, role-based security, and a grounded AI assistant - architected as a decoupled frontend and backend from the outset.
02. Data Model and Backend Core
The team designed the relational schema in SQLAlchemy, covering users, issues, sprints, worklogs, active timers, invitations, and the AI-related document and session tables, then built out full CRUD endpoints for issues, subtasks, worklogs, and sprints under FastAPI's async architecture.
03. Authentication and RBAC
JWT-based authentication and role-based access control were implemented to distinguish managers from employees, gating sensitive actions like sprint creation and invitations behind manager-level permissions.
04. Frontend Board and Time Tracking
The team built the drag-and-drop Kanban board, backlog view, global timer header, and task detail drawer, wiring frontend state to backend timer and worklog endpoints so time tracking stayed accurate across sessions.
05. Time Analytics
The analytics dashboard was built to surface logged-vs-estimated hours, priority distribution, and recent activity, giving managers a real-time view of team performance.
06. Three-Layer AI System
The team implemented the core AI service with Groq integration and a local embedding fallback, built the bounded-tool agent for safe natural-language actions against backend data, and developed the RAG pipeline for document ingestion, chunking, and grounded query answering.
07. Testing and Refinement
The full application - frontend and backend - was tested across core workflows: issue and sprint management, time tracking accuracy, invitation and onboarding flows, and AI agent/RAG response accuracy against real backend data and ingested documents.
Key Features / Deliverables
- Drag-and-drop Kanban board with To Do, In Progress, Review, and Done columns
- Inline play/stop time tracking directly on issue cards
- Backlog view with filtering, quick-add, and update/delete controls
- Global sticky timer header with play/pause/stop and notes logging
- Jira-style task detail drawer with subtask checklist and worklog timeline
- Real-time time analytics dashboard (logged vs. estimated hours, priority distribution, activity feed)
- Full CRUD sprint and issue management with employee sprint assignment
- JWT-based authentication with manager/employee role-based access control
- Role-scoped email invitation and onboarding system
- Three-layer AI architecture: provider-agnostic AI service, bounded-tool AI agent, and RAG-based document query system
- Document ingestion and semantic chunking for grounded, policy-accurate AI answers
- Multi-turn AI conversation history via dedicated session and message tables
Challenges & Solutions
Building a platform that merges real-time collaboration tooling with a safe, grounded AI layer introduced several technical challenges the Meritorious CodeCrafters team had to design around:
Keeping AI actions safe and bounded. Letting an AI agent act on real backend data - updating issues, fetching tasks, generating invitations - carries risk if left unconstrained. The team built the AI agent around a bounded set of defined tools rather than open-ended database access, so natural-language requests translate into specific, predictable backend transactions.
Grounding AI answers in the team's actual documentation. Generic LLM responses aren't reliable for internal policy questions. The team built a dedicated RAG pipeline that ingests and chunks uploaded documents into embeddings, so the AI Agent answers policy and onboarding questions strictly from ingested content rather than general model knowledge.
Avoiding hard dependency on a single AI provider. To keep the AI layer resilient, the team built a provider-agnostic core AI service that routes to Groq for primary processing but falls back to a deterministic, dependency-free local embedding algorithm for similarity search, reducing reliance on external AI infrastructure for core retrieval functionality.
Keeping time tracking accurate across a real-time, multi-user environment. Timer state needed to stay consistent whether a user was starting, pausing, or stopping a tracker, and that state needed to convert cleanly into permanent worklog records. The team built dedicated timer endpoints handling start, pause, and stop operations per user, with automatic conversion into worklog entries once a timer session ends.
Enforcing role-based permissions consistently across a growing feature set. With sprint creation, invitations, and reporting all needing different access levels, the team implemented JWT-based RBAC at the endpoint level, so restricted actions are consistently blocked for non-manager roles across the entire API surface.
Results
Meritorious CodeCrafters delivered a fully built, decoupled full-stack platform giving the client's users:
- A unified Kanban, sprint planning, and time-tracking workspace in place of multiple disconnected tools
- Native, per-issue time tracking with automatic worklog generation, eliminating manual time entry
- A three-layer AI system capable of safely acting on real backend data and answering questions grounded in the team's own documentation
- Role-based security enforced consistently across sprint management, invitations, and reporting
- A streamlined onboarding flow through role-scoped invitations
- A resilient AI architecture that isn't fully dependent on a single external AI provider
Business Impact
By unifying issue tracking, sprint planning, time tracking, and AI-assisted knowledge access into a single platform, AeroFlow reduces the tool-switching and manual overhead that typically slows down agile software teams. The bounded AI agent gives developers a faster way to interact with their own tasks and status updates, while the RAG-grounded assistant turns static internal documentation into something developers can actually query in the flow of their work - shortening the distance between "I have a question" and "I have an accurate answer." The modular, three-layer AI design also gives the client a resilient, extensible foundation that isn't locked into a single AI provider, positioning the platform to evolve its AI capabilities over time.
Why Meritorious CodeCrafters
Meritorious CodeCrafters brought together full-stack engineering and applied AI architecture to deliver a platform that goes well beyond a typical issue tracker. Designing a bounded, tool-based AI agent alongside a dedicated RAG pipeline - while keeping the system provider-agnostic and resilient - reflects the AI-first engineering approach Meritorious CodeCrafters applies across its custom software and AI development work, particularly for teams building sophisticated, data-grounded AI features rather than surface-level chatbot integrations.
Final Outcome
AeroFlow stands as a comprehensive example of how agile team tooling can be extended with a genuinely useful, safely bounded AI layer. Meritorious CodeCrafters delivered a decoupled, modern full-stack application - spanning Kanban and sprint management, native time tracking, role-based security, and a three-layer AI and RAG architecture - giving the client a strong, extensible foundation for the platform's continued evolution.
Call to Action
Looking to build an AI-assisted platform grounded in your own operational data? Contact Meritorious CodeCrafters to discuss your AI development or custom software project.
Suggested FAQs
What problem does AeroFlow solve?
AeroFlow unifies Kanban-based issue tracking, sprint planning, time tracking, and AI-assisted knowledge access into a single platform, replacing the disconnected tools teams typically juggle for agile delivery.
How does AeroFlow's AI assistant interact with real project data?
AeroFlow uses a bounded-tool AI agent that translates natural-language requests into specific backend actions - like fetching personal tasks or updating issue status - rather than allowing open-ended access to the database.
How does AeroFlow ensure AI answers about company policy are accurate?
AeroFlow includes a dedicated RAG (retrieval-augmented generation) system that ingests and chunks uploaded documentation into embeddings, so AI answers are grounded strictly in the team's own ingested documents rather than general AI knowledge.
Does AeroFlow depend on a single AI provider?
No. AeroFlow's core AI service is provider-agnostic, routing primary requests to the Groq API while falling back to a deterministic local embedding algorithm for similarity search when needed.
How does time tracking work in AeroFlow?
Users can start, pause, and stop timers directly from issue cards or a global sticky timer header. Completed timer sessions are automatically converted into persistent worklog entries tied to the relevant issue.
How does AeroFlow handle team roles and permissions?
AeroFlow uses JWT-based authentication with role-based access control, distinguishing managers from employees and restricting sensitive actions like sprint creation and invitations to manager roles.