Users can ask questions across a collection instead of learning a search syntax or manually opening every file.
Project portfolio
Piotr Paklerski Projects
Working Gen AI and data-product demos focused on enterprise use cases: natural-language interfaces, governed retrieval, evidence-backed answers, and decision surfaces that are usable by non-technical stakeholders.
Retrieval over real documents
Verified citations
Human-in-the-loop controls
Enterprise Document Intelligence
How many cyber-related observations were reported in 2025?
✓
Route from verified index
Read selected reports · verify citations · label caveats
What this demonstrates
Enterprise AI patterns, shown in a working browser app
Factual claims are tied back to source passages, and unsupported or unverified citations are visibly quarantined.
Counts, filters, and dashboard figures come from a verified index; the model is used for language understanding and synthesis.
The demo uses a limited built-in proxy by default, while still supporting user-provided Claude or open-source model endpoints.
Featured
Live projects
Gen AI · document intelligence · public-data demo
Enterprise Document Intelligence
Natural-language interrogation over 39 public UNICEF OIAI reports, with deterministic dashboards, scoped retrieval, verified verbatim citations, and a limited built-in Haiku proxy for reviewers.
- Routes questions to relevant reports before reading full text.
- Separates exact counts from model-generated narrative synthesis.
- Shows how AI outputs can be constrained, cited, and reviewable.
39 reports
280 observations
651 verified quotes
Next
More demos will live here
This page is structured as a portfolio hub, so future apps can be added without changing the share link.
Ask the reports
All reports
Default demo API: Claude Haiku via a limited proxy. It may be less capable or temporarily unavailable; use Settings for your own Claude key or open-source endpoint.
Browse
Report library
Loading report text…
Questions are answered only from this report. Every factual
claim carries a verified verbatim citation; anything the report doesn't cover is answered with
“Not covered in this report.”
Ask across all reports. Counts come from the verified
index; narrative claims carry verified verbatim citations. The default demo API uses Claude Haiku through a limited proxy; switch provider in Settings for stronger or private usage.
Route from indexRead selected reportsVerify citations