Jovan OS Lite
A local agentic planning and review system that connects goals, weekly tasks, evaluator feedback, weekly review, and human-approved optimization into one structured workflow for tracking progress across study, AI projects, sport, and career development.
Scattered goals and weak review loops
Personal goals often live across notes, habits, tasks, and vague intentions, making it hard to see what is actually moving forward.
Planner -> evaluator -> review -> optimizer
The system turns goals and weekly actions into structured plans, evaluates progress, creates review summaries, and suggests changes for human approval.
Clearer execution across multiple domains
The goal is to make progress easier to inspect, rebalance, and improve without relying only on motivation or memory.
Personal goals and weekly reviews are hard to keep consistent without state, evaluation, and structured recommendations.
Shows the ability to build multi-step agentic systems with state, evaluation, recommendations, and human approval.
- Planner and evaluator agents
- Weekly review and optimizer loop
- Human-approved recommendations
- SQLite persistence
- Gradio interface with structured outputs
From input to a reviewable next action.
- 01
Load goals and weekly context
Bring selected goals, priorities, and recent progress into one reviewable workflow.
- 02
Generate a practical plan
Prepare focused actions for the week based on the active goals and current constraints.
- 03
Evaluate progress
Compare completed work, consistency, and momentum against the intended direction.
- 04
Create weekly review
Summarize what moved forward, what stalled, and what needs attention next.
- 05
Suggest optimizations
Recommend weight or priority changes, but keep final decisions human-approved.
Case study notes
This project started from a personal execution problem: when several important domains move at the same time - university, AI projects, sport, career visibility, and long-term planning - it becomes easy to confuse activity with progress.
Instead of building a generic to-do app, I designed Jovan OS Lite as a local agentic workflow. The system helps prepare plans, evaluate progress, generate weekly reviews, and suggest optimizations while keeping the final decision visible to the user.
The goal of the prototype is not to automate life decisions. The goal is to make priorities easier to inspect, progress easier to review, and weekly execution easier to improve.
Visible proof, without private operational data.
1 / 7
Tools, concepts, and architecture
The workflow is designed as a sequence of small AI-assisted steps rather than one large autonomous life-management agent.
- 1Load goals and weekly context.
- 2Generate a practical weekly plan.
- 3Evaluate progress and consistency.
- 4Create a weekly review summary.
- 5Suggest optimizations for human approval.
Detailed case study report
For visitors who want to understand how the project was shaped, how I tested it, and how I think about agentic workflow design.
Why I built it
I wanted a practical way to manage several important domains at once without relying only on scattered notes, motivation, or memory. The project became a testbed for applying agentic AI patterns to personal execution.
Problem framing
The core problem is not simply planning tasks. It is knowing which goals deserve attention, whether progress is actually happening, and when priorities need to be rebalanced.
Workflow design
I shaped the system around a loop: define goals, generate weekly actions, evaluate progress, review the week, and suggest optimizations that still require human approval.
AI role
The AI role is limited and practical: structure plans, evaluate outputs, summarize progress, and prepare recommendations. It is not positioned as a fully autonomous decision-maker.
Testing approach
I tested whether the system produced useful, inspectable outputs: clear plans, understandable evaluations, meaningful weekly reviews, and optimization suggestions that could be accepted or rejected.
Next improvements
The next layer would improve the dashboard experience, add better historical trends, support richer goal analytics, and make the review process easier to use over longer periods.
What this proves
Agentic workflow structure
A useful agentic system does not need to be fully autonomous. Planner, evaluator, review, and optimizer loops can create value while staying reviewable.
Human-approved optimization
The strongest part of the design is the approval step: the system can suggest changes, but the user stays responsible for final priority and weight decisions.
What I would improve next
I would improve long-term tracking, make the dashboard more visual, add better trend analysis, and refine how the optimizer explains its recommendations.
Want to discuss an agentic workflow like this?
Send me a message if you are exploring planning, evaluation, review, or human-approved AI workflows.
