Sprints planned by math, not vibes.
Three-phase calibration - historical velocity, tenure, component expertise. Then an 8-factor scoring pass across your backlog. You read every weight.
The problem
Sprint planning is a guess dressed up as a ceremony.
- You don't know your team's real velocity. You know the number you quoted product last quarter.
- You don't know who has slack this sprint. Utilization is a gut check, not a grid.
- You don't know if a task fits a developer. You know who has time.
- Most AI planners 'help' by ingesting every ticket into a model and generating a recommendation you can't audit. That fixes the speed problem and creates a security problem.
How it works
Step by step.Nothing behind a curtain.
Phase 1 - Velocity
We pull up to 10 past sprints and compute per-team and per-developer velocity. Coefficient of variation under 0.3 = 'consistent.'
Phase 2 - Tenure
HiBob gives us months at company. We derive experience bands: <12 months junior, 12–36 mid, 36–72 senior, 72+ expert.
Phase 3 - Component expertise
ComponentHistory tracks per-developer which components they've touched, how often, for how many story points.
Scoring
Each backlog task gets an 8-factor weighted score: priority, backlog age, labels, epic alignment, story points, dependencies, business value, skill match. Every weight is 0–5 and configurable.
Selection
Greedy fill within capacity. 85% buffer by default. 10% overflow allowance for high-scoring small tasks.
Output
Plan with goals, warnings, confidence rating, Redis cache for insights and follow-up Q&A.
What you get
Specific.Not marketing-department specific.
- 4 presets out of the box: balanced, deadline, quality, velocity.
- Capacity grid with per-developer current capacity (historical if they have ≥ 3 sprints, default 13 SP otherwise), assigned SP, available SP, utilization percent.
- Warnings when the plan is < 70% filled ('underfill') or contains unresolved dependencies.
- Confidence rating: high (≥ 3 sprints of history AND ≥5 tasks), medium, low.
- Narrative insights streamed live. Ask follow-up questions on the plan; the answers stream over Server-Sent Events.
Built right
Math you can read.Plans you can audit.
Every weight in the scoring pass is configurable and visible. Plans show the reasoning, not just the output - change a weight, re-score the backlog, see the new plan instantly. The full architecture is something we'd rather walk you through on a call.
Get the architecture deep-dive on a callWorks with
Plan the sprint. Read every weight.
Private by default. Specific by design. Free for 2 users, forever.