Keep financial calculations reproducible.
I separated exposure and scenario calculations from AI interpretation, keeping the baseline inspectable.
See the exposure example ↗Portfolio AI
From a list of holdings to a clearer understanding of what you own—and how a decision changes it.
Independently conceived and built by Ashraf Vallur.
Product strategy, AI architecture, and AI-assisted engineering.
AI interpretations shown: supervised local tests.
A holdings list separates NVIDIA from QQQ, VOO, and VTI. Their underlying company exposure can overlap.
The product connects those ownership paths, makes the known exposure inspectable, and shows how a hypothetical change affects it.
A $545,450 synthetic portfolio · 7 accounts · 22 positions.
Move from visibility to a hypothetical decision without skipping the evidence.
Bring accounts, holdings, and allocations into one view. Establish the financial context for the observations and scenarios that follow.
Synthetic data · fixed snapshot
Rocket Lab’s known exposure exceeds the synthetic investor’s 7% company guideline. The observation connects its explanation to inspectable portfolio facts.
The guideline is a synthetic input. The observation prompts investigation; it does not recommend a transaction.
Make the interpretation reviewable, with the supporting facts close at hand.
Readable editorial summary of the supplied local-test capture.
Open original Rocket Lab capture (opens in a new tab) ↗
Local interpretation evidence is distinct from hosted availability. Reference checks do not establish semantic correctness.
One company. Direct shares and multiple ETF ownership paths. A reproducible calculation makes the known exposure visible.
Known exposure reflects supplied constituents. Partial ETF coverage can leave additional exposure uncounted.
Open original NVIDIA capture (opens in a new tab) ↗
A hypothetical $20,000 RKLB → VTI move reduces one holding while adding indirect ownership of another. The total portfolio value stays fixed.
| Holding / exposure | Before | After |
|---|---|---|
| RKLB | $59,200 | $39,200 |
| VTI | $68,400 | $88,400 |
| Known NVIDIA | $42,244 | $43,244 |
| Portfolio total | $545,450 | $545,450 |
Fixed illustrative 5% NVIDIA weight in VTI. Excludes fees, taxes, slippage, and market movement. Hypothetical analysis; no recommended or executed trade.
Case-study illustration; no application or model API calls. VTI uses a fixed 5% NVIDIA weight; RKLB contributes no NVIDIA exposure. Portfolio value stays $545,450.


Product judgment becomes concrete in what the system calculates, what it explains, and what remains useful when interpretation fails.
I separated exposure and scenario calculations from AI interpretation, keeping the baseline inspectable.
See the exposure example ↗I connected observations to exact evidence references. Those checks support review; they do not establish semantic correctness.
See the observation ↗I kept deterministic scenario results available when interpretation is unavailable. The explanation narrows; the calculation remains accessible.
See the scenario ↗Investor intent and final decisions remain human responsibilities.
Three historical experiments connect development decisions to product judgment. Explore the supporting depth behind the executive story.
In a reported development experiment, a useful AI response was rejected because its evidence references failed validation. The product decision was to improve the AI contract while preserving the validator.
/snapshot/top_holdings/0/portfolio_weightRequired path/top_holdings/0/portfolio_weightThe design moved company-level exposure into deterministic calculations rather than asking the model to infer ETF composition. The NVIDIA example above shows the supplied financial baseline.
Return to the exposure example ↗The historical experiment compared known company exposure with an explicit synthetic guideline. A known exceedance and a below-guideline value with incomplete coverage support different conclusions.

Portfolio visibility, company-level exposure, a hypothetical scenario, and useful deterministic fallback.
AI interpretations are documented in supervised local tests, separately from the hosted experience.
Synthetic data. Partial ETF coverage. Exact reference checks do not prove semantic correctness. No execution or investment recommendations.
More scenarios, missing-data handling, semantic evaluation, provider resilience, and production AI readiness.
Market & Regime Intelligence and Opportunity Intelligence remain product directions, not demonstrated capabilities or delivery commitments.
Portfolio analytics and hypothetical scenario analysis, with inspectable exposure insights.
Explore Live Demo (opens in a new tab)Hosted demo: portfolio analytics and scenario analysis. AI-generated interpretations are demonstrated separately.
Portfolio analytics and hypothetical scenario calculations are deterministic. AI interpretation is optional and subject to server configuration and request limits. The AI interpretations shown in this case study were captured during supervised local tests. Rule-based observations are supported separately. No transactions are executed.
Evidence-reference checks validate paths and values against supplied facts. They do not establish overall semantic accuracy. Original screenshots remain unchanged, including their documented capture boundaries.