AI in the Real World

AI that solves the boring, expensive parts.

PTRS doesn't use AI as a novelty — it uses it for the high-volume, low-joy tasks: reading paper forms, drafting parent messages, answering data questions, classifying incidents, translating for families, and finding patterns in mountains of records. Always with PII protection built-in.

How it stays safe

Children's data never leaves your organization.

Every AI request goes through a mandatory 7-step pipeline. Names, SSNs, DOBs, phones, emails, and addresses are tokenized before a single word reaches the model — and any response is scanned for leaks before it's ever shown to a user.

The mandatory 7-step AI pipeline · COPPA · FERPA
1 · Tokenize Names → [NAME_001] SSN → [SSN_003] 2 · Route Fast 8B or Complex 70B 3 · Build prompt + RAG context (if regulation) 4 · Call LLM Ollama (local) or Claude (cloud) 5 · PII scan Any leak? → [REDACTED_*] 6 · Detokenize [NAME_001] → real name (client-side) 7 · Audit log Tokenized only (never raw PII) User query → → Clean, detokenized response
🏠

Local-first

Default provider is Ollama running on the org's own hardware. No data leaves the network.

🛡️

Two-layer defense

Tokenize before the model sees anything. Scan after in case something slips through.

SQL safety validator

AI-generated SQL must be SELECT-only against a 32-table whitelist before it ever runs.

Real scenarios · Real people

Eight ways AI earns its keep at a club.

Staff Member1 · Reading a paper enrollment form
A family showed up at the front desk with a handwritten enrollment packet. I need to get it into the system before lunch.

What PTRS does: Staff opens Quick Scan, snaps a photo of the form. Claude Vision extracts every field — child name, DOB, guardians, emergency contacts, allergies, medical info — into a structured JSON with confidence scores. Staff reviews, corrects anything flagged, and submits. A 20-minute data-entry job becomes a 90-second review.

Powered by: AI Task EntityExtraction · multi-page batch support · confidence thresholds · review-before-commit UI.

Site Director2 · Asking a plain-English question about the data
How many kids in our after-school program had more than 5 absences this semester?
DIR
How many kids in our after-school program had more than 5 absences this semester?
AI
47 children in the After-School program had 5+ absences. Top patterns: illness (23), family appointments (19), transportation (5). I'd recommend outreach to the 8 members with 10+ absences — want me to generate a list?

Powered by: NlToSql (llama3.3:70b, temp 0.0) · 32-table SELECT-only whitelist · results scanned for PII before display.

Floor Staff3 · Writing an incident report under pressure
A child just had an allergic reaction. I have two minutes to document everything before the parents arrive.

What PTRS does: Staff opens the 5-step incident wizard. The AI Wizard Panel on the right analyzes each step as it's entered — suggesting a severity level, flagging whether this triggers OCCL's 24-hour notification, and generating a structured markdown narrative when the staff member is done. A branded PDF is ready to export in the same minute.

Powered by: IncidentClassification (fast 8B) · IncidentStructure (complex 70B) · deadline tracker · QuestPDF export.

Parent Liaison4 · Sending a notification in three languages
An emergency action plan was activated. I need to tell every guardian, in their preferred language.

What PTRS does: The notification dispatcher generates the base message, then the AI translates into Spanish and Haitian Creole — preserving the formal tone required for an EAP activation. Each guardian receives it in the language set on their portal profile, via the channel they chose (SMS, email, push), honoring quiet hours.

Powered by: Translation (fast 8B, temp 0.1) · i18n locale preferences · multi-channel dispatcher with retry.

Compliance Officer5 · Interpreting a DELACARE regulation
What is the required staff-to-child ratio for mixed-age school-age groups under DELACARE?
CO
What is the required ratio for mixed-age school-age groups under DELACARE?
AI
According to the DELACARE §49.7 rule indexed in your regulation library, mixed-age school-age groups (ages 5–12) require a 1:15 ratio when the youngest child is at least 6 years old. If the group includes a 5-year-old, the more restrictive 1:12 applies. (Source: DELACARE Regulations, §49.7, page 87.)

Powered by: RegulationQA (complex 70B) · RAG retrieval over pgvector-indexed regulations (top-5 · 512-token chunks with 64-token overlap).

Program Director6 · Brainstorming a new summer program
Give me a skeleton for an 8-week summer AI / robotics program for ages 10–14.

What PTRS does: The program builder's AI panel drafts: a charter statement, 8 weekly learning objectives, suggested curriculum units, pre/post assessment ideas, and a short list of potential community partners. Director keeps what works, edits what doesn't, and saves the program — no blank-page paralysis.

Powered by: ProgramIdeaGeneration · CurriculumOutlineGeneration · GoalKpiSuggestion · AssessmentDesign · PartnershipSuggestion (all 70B).

Safety Officer7 · Finding patterns in 200 incident reports
We had more incidents in the last 90 days than the same period last year. Where should we focus?

What PTRS does: AI pattern-analysis groups incidents by location, time-of-day, type, and staff-on-duty. It finds that Wednesday afternoons in the Wilmington gym account for 34% of minor injuries and correlate with one specific rotation. The dashboard ships that insight with a clickable drill-down.

Powered by: PatternAnalysis (70B) over tokenized incident data · feeds the Incident Trend 3-panel chart.

Finance / ED8 · Generating a funder report
United Way wants a Q1 report by Friday — attendance, demographics, outcomes, and a narrative summary.

What PTRS does: Finance picks United Way from the funder CRM, selects Q1 2026, and clicks Generate. PTRS aggregates the numbers from the live database; AI drafts a 3-paragraph narrative summary and recommendations. Finance reviews, edits the narrative, exports the PDF, and logs the submission to the funder record — all inside an hour.

Powered by: FunderReport (70B, temp 0.2) · funder CRM · PDF generation · audit log of submission.

All 23 AI tasks in the system

Where AI shows up — by job, not by demo.

TaskModelWhat it does
NlToSql70BConverts English questions into safe SELECT queries
RegulationQA70B · RAGAnswers DELACARE / OCCL / USDA questions from indexed docs
ComplianceReport70BGenerates the compliance summary narrative
IncidentStructure70BFormats a free-text incident description into a structured report
IncidentClassification8BSuggests severity (1–5) + OCCL / BGCA flags
PatternAnalysis70BFinds recurring locations / times / types in incident data
GrantReport / FunderReport70BDrafts grant narratives from live program data
BoardReport70BMonthly board-level narrative summary
WeeklySummary70BSite-level weekly operational summary
AttendanceSummary8BNatural-language recap of attendance trends
EntityExtraction8B · VisionExtracts fields from a photo of a paper form (Quick-Scan)
Translation8BEN ↔ ES ↔ HT translation of parent-facing text
ParentNotification8BDrafts parent-facing notification messages
SentimentAnalysis8BLight sentiment scoring on messaging and survey free-text
MenuCategorization8BAuto-categorizes food items for the CACFP meal-pattern check
ProgramIdeaGeneration70BBrainstorms new program concepts
CurriculumOutlineGeneration70BDrafts units, sessions, activities for a curriculum
GoalKpiSuggestion70BSuggests goals and measurable KPIs for a program
SuccessCriteriaSuggestion70BDefines what "success" looks like for a program
AssessmentDesign70BDrafts pre / post assessments aligned to program goals
PartnershipSuggestion70BSuggests community / corporate partners for a program
ComplianceGapAnalysis70BIdentifies the most material gaps before an OCCL inspection
The point

AI as a staff member, not a gimmick.

PTRS's AI does the tasks that used to take after-hours: reading a form, drafting a report, answering "how many…" questions, writing a parent message in the right language, and finding patterns nobody has time to look for.

And it does all of that without children's personal data ever leaving your organization.

See the business flow Browse all modules