The children's record
Six months of messages, documents and receipts, organized and traced to their sources.
Sources312fingerprinted originals
Handoffs on time94%47 of 50
Avg. care follow-up23dreferral to appointment
Shared costs$4,210paid with receipts
Messages per month
Jan
Feb
Mar
Apr
May
Jun
What the record shows
| Median reply time, Parent A | 3.4h |
| Median reply time, Parent B | 27h |
| Reimbursement requests answered | 6 / 9 |
| School events attended by both | 4 / 7 |
Gap finder: three records would strengthen this file. Review them →
Timeline
Every entry links back to the original it came from. Click one to see its source.
Select an entry to see the source document and its fingerprint.
Add records
Drop in anything: PDFs, text exports, emails, photos. Each original is fingerprinted, dated, and tagged with the people in it.
Drop files here
This demo doesn't upload anything. Use the sample batch to watch the pipeline.
Processing log
| Nothing processed yet. |
Gap finder
The record mentions these documents but doesn't contain them. Request them before someone asks why they're missing.
Brief builder
Pick the topics. Get a neutral, footnoted summary a professional can verify line by line.
How it's built
The pattern behind this demo, which I've shipped in a live product for families.
- Ingestion. Mixed file types are normalized to text plus metadata. The untouched original is hashed on arrival so any later excerpt can be proven against it.
- Extraction. An LLM pulls dated events, people, amounts and commitments into a fixed schema. Anything it can't ground in a quoted span is dropped, not guessed.
- Entity resolution. "Dr. P", "the pediatrician" and "Dr. Patel" collapse into one person the user confirms once.
- Pattern analytics. Plain arithmetic over the extracted events: on-time rates, reply medians, follow-up delays. No model opinions in the numbers.
- Gap detection. References to documents that never arrived become a request list.
- Privacy. Private by default, scoped sharing, no training on user data.