Personalized Future Planning Platform

People make major life decisions — finances, career, family — with almost no way to see the downstream consequences of the trade-offs. The product turns messy, personal inputs into simulated future pathways and a structured report someone can actually act on.

AI product engineer. I designed the data model, the simulation and recommendation layers, and the report-generation pipeline.

1. 01Questionnaire — structured capture of financial, lifestyle, and family inputs.
2. 02Structured Profile — validated, typed representation of the user's situation and goals.
3. 03Simulation Engine — projects future pathways under different assumptions and choices.
4. 04Recommendation Layer — synthesizes simulations into prioritized, explained guidance.
5. 05PDF Report — a structured, shareable decision document.

• Structured profile as the contract

  A validated typed profile sits between free-form input and the engine, so the simulation always receives clean, well-formed data and the LLM stages stay grounded in real fields.

• Simulation before generation

  Numbers come from an explicit simulation engine, not from an LLM. The language model narrates and explains results it did not invent, which keeps the output trustworthy.

• FastAPI service boundaries

  Clear API boundaries between questionnaire, simulation, and report generation keep each stage independently testable and replaceable.

Validated by checking simulation outputs against hand-computed scenarios, and by reviewing generated reports for internal consistency and faithfulness to the underlying numbers. Report quality is assessed on clarity, correctness, and actionability.

• Garbage-in from ambiguous questionnaire answers propagates; input validation and sensible defaults reduce but do not eliminate this.
• LLM-written narrative can drift from the simulated numbers if not tightly grounded; the report pipeline pins claims to computed values.
• Long-horizon simulations compound assumption error; the report communicates uncertainty rather than implying false precision.

• Python
• FastAPI
• Simulation
• LLMs
• Report Generation