A multi-agent product management copilot built with NVIDIA Nemotron + LangGraph
Automating the product management lifecycle — from messy customer feedback to Jira backlogs and CI/CD plans.
Sprintly was born from watching real product managers spend countless hours on coordination instead of creativity.
We saw the same pattern over and over:
- Reading long research reports and survey data
- Manually writing Jira epics, stories, and acceptance criteria
- Constantly syncing with DevOps and engineering to plan releases
When we saw a hackathon track focused on product manager workflows, we decided to build an AI system that acts like a digital PM partner — one that researches, plans, and coordinates execution end-to-end.
At the same time, we were experimenting with NVIDIA Nemotron models and were impressed by their reasoning and planning capabilities. That led to Sprintly: a multi-agent system orchestrated by LangGraph that thinks and works like a PM.
Sprintly is a connected workflow assistant for product managers.
From idea → research → Jira backlog → DevOps plan, all in one continuous flow.
Model: Nemotron Nano 9B v2
- Ingests customer feedback, survey data, and research docs
- Summarizes key trends and pain points
- Surfaces product opportunities and passes structured insights downstream
Why it’s useful: PMs get instant research summaries and structured insights instead of digging through raw data.
Model: Llama 3.3 Nemotron Super 49B v1.5
- Converts ResearchAI outputs into epics, user stories, and acceptance criteria
- Maintains linking between epics → stories → priorities
- Integrates with Jira Cloud API to automatically create and update issues
Why it’s useful: Removes the manual grunt work of writing Jira tickets so PMs can focus on direction and scope.
Model: Nemotron Mini 4B Instruct
- Scans the repository (tests, package files, existing workflows)
- Proposes CI/CD workflows, sprint plans, and release checklists
- Can generate
.github/workflows/*.ymland open pull requests via GitHub REST API
Why it’s useful: Bridges the gap between planning and engineering, helping PMs connect strategy → deployment.
Sprintly uses LangGraph as a routing and memory layer:
- Routes each user request to ResearchAI, JiraAI, DevOpsPlannerAI, or a combination
- Shares state between agents so output from one is clean input to the next
- Supports flows like:
ResearchAI → JiraAI → DevOpsPlannerAI
(insights) (backlog) (CI/CD & sprint plan)