The question behind the product
Operations teams must maximize useful output within finite time, capacity and resource constraints. Logistics routes, production schedules, sales beat plans and field-service assignments all depend on mandatory tasks, sequence, geography and changing priorities. Powerful planning systems can still be difficult for people to use. PujoPlan AI explores whether natural-language interaction can simplify these decisions without letting a language model invent operational feasibility.
The hypothesis
An AI layer can translate human intent into structured planning constraints, while a deterministic engine remains responsible for calculation, validation and execution rules. The product pattern is transferable; the consumer experiment demonstrates it without claiming enterprise deployment or measured enterprise outcomes.
The product approach
Separate interpretation from feasibility. AI understands requests such as reducing walking, preserving essential stops or finishing earlier. Structured changes are passed to the planning engine. Hard constraints remain distinct from preferences, and a fluent explanation never substitutes for a feasible plan.
Enterprise parallels include a logistics manager adding urgent deliveries, a production planner reducing changeovers without delaying priority orders, a sales manager redistributing mandatory outlet visits and a service team adjusting technician schedules. These are possible applications of the design pattern, not features claimed to be deployed in this experiment.
How it works
1. Capture the planning brief: date, starting location, available time, finish time, essential pandals, walking tolerance and group mobility needs.
2. Translate intent: AI turns natural-language requests into structured preferences and constraints, rather than independently choosing a plausible-sounding route.
3. Use curated operational data: locations, geographic clusters, nearest Metro stations, viewing allowances and accessibility context support planning. Unknown information stays unknown; AI should not fabricate queue times, access restrictions or entrance details.
4. Optimize and validate: the deterministic engine evaluates route sequence, travel and available time while respecting mandatory stops. If priorities conflict, the product should surface the trade-off rather than silently discard an essential stop.
5. Replan during execution: requests such as “we are 40 minutes late” or “reduce walking” update structured inputs and recalculate the remaining plan.
6. Explain the outcome: present a practical itinerary and its trade-offs in language a user can act on. The same interaction pattern can sit above ERP planning engines, transport systems, PJP tools and field-service scheduling systems.
What this work reinforces
The hardest product decision is where AI responsibility ends. Interpretation and orchestration suit the AI layer; distance, time, capacity and mandatory rules belong to structured data and deterministic calculations.
Evaluate feasibility separately from fluency. Relevant evaluation criteria include mandatory-stop retention, finish-time compliance, travel burden, constraint violations, replan consistency and clarity of trade-off explanations. These are evaluation dimensions, not published performance results.
For enterprise product management, the opportunity may be to simplify how users express business needs to existing planning engines rather than replace those engines with an unconstrained language model.
