What You Can Buy Today — and What Is Coming
Every item on this page carries one of three labels, so you know exactly what you are getting.
Available now Running in production or in our reference deployment today.
Built to order We engineer it for you as a scoped project, using components we already run.
In development On our roadmap. Join the waitlist; do not plan a launch around it yet.
Featured · 01
Document-to-Knowledge Pipeline: OCR, Chunking, Embeddings and a Local Model
Turn scanned and digital documents into searchable, de-identified knowledge on hardware you
control. The pipeline reads documents with OCR, removes personal identifiers, splits and embeds the
text, and answers questions from your material with a language model that runs on your servers.
1CapturePhone, scanner, camera or upload
2OCRImage and PDF to text
3De-identifyRemove personal identifiers
4ChunkSentence-aware splitting
5Embed1024-dimension vectors
6RetrieveVector search with source checks
7AnswerLocal LLM, grounded in your sources
What runs today Available now
- Image and PDF OCR with a confidence-gated cascade of open OCR engines, HEIC and RAW photo conversion, and a face-blur step before OCR.
- Sentence-aware chunking, and local embeddings served from your own hardware.
- Vector retrieval on PostgreSQL with pgvector, with per-question source selection and an abstain-when-unsure rule.
- Answers checked against the retrieved sources by a groundedness judge; a crisis-language screen runs before generation.
- Open-weight models behind a single OpenAI-compatible local gateway; the gateway does not log prompt or response content.
- In our reference deployment, service-to-service traffic stays on the local machine and no third-party AI API is called from any request path.
How you can deploy it
- Self-hosted on your servers Available now — we size the GPU hardware, install, harden and hand over a runbook. We test the deployment with outbound network access blocked as part of the engagement.
- Your own model Built to order — bring a fine-tuned model and we integrate it, or we train adapters on your data. Our own adapters are trained with the same methods.
- Edge capture Built to order — cameras, scanners or phones that capture on-site and send images into the pipeline, with permissively licensed OCR that can run on the device.
- Your cloud tenancy In development — container and Helm packaging for deployment inside your own cloud account.
- Metered API access In development — keys, per-key limits and usage billing. Join the waitlist.
OCRChunkingEmbeddings
pgvectorRAGOpen-weight LLMs
Self-hostedEdge capture
You receive
A working pipeline on your infrastructure, tuned to your document types and vocabulary, with an evaluation
baseline on your own sample documents and a runbook your team can operate.
Accuracy depends on your documents. We benchmark on your samples before you commit; we do not quote a general OCR accuracy figure.
Featured · 02
De-Identification and Scrubbing Software
Detect and remove personal identifiers from clinical and business text before it reaches storage, search or a
model — self-hosted, with published recall by category and honest limits.
What runs today Available now
- Built to detect all 18 identifier categories listed in 45 CFR 164.514(b)(2)(i), combining a clinical-text NER model with pattern recognizers for names, dates, contact details, account and record numbers, URLs, IP addresses and more.
- Keeps clinical codes (ICD-10, CPT) and lab values in the text so the result stays useful.
- Redacts all dates by default; ages over 89 are reported as 90+.
- Reads text, PDFs and scanned images through the OCR path above.
- Runs on your hardware; text is not sent to a third-party API.
Coming to the product In development
- Date-shifting mode — a consistent, per-subject date offset so the order and spacing of wearable, step-count, continuous-glucose-monitor and lab timestamps survive de-identification.
- Pseudonymous subject and device identifiers so external data streams can be joined without exposing the person.
- Structured input (CSV, JSON, FHIR), batch requests and entity-level results with confidence scores.
- Per-request policy profiles and sector packs (legal, HR, insurance claims, finance, education).
- Measured recall for every identifier category, including those not yet benchmarked.
Measured recall on a synthetic healthcare benchmark (Nemotron-PII Healthcare, 13,808 records, June 2026)
| Identifier | Recall |
|---|
| Names | 96.4% |
| Medical record numbers | 97.2% |
| Phone numbers | 99.3% |
| Email addresses | 99.9% |
| Social Security numbers | 99.4% |
| URLs | 100% |
| Device identifiers | 87.8% (41 examples) |
These figures are recall (the share of identifiers removed), measured on synthetic data, with dates and
ages 89 or under excluded from scoring. They are not a guarantee of results on your data, and precision
is not reported here. Categories without a published figure have not yet been benchmarked. Date-shifted
data is not de-identified under the HIPAA Safe Harbor method; it is typically used under an Expert
Determination or a limited data set with a data use agreement, and we recommend that review. NexGenHealth software does not by
itself make an organization compliant with any law.
Featured · 03
MAHA-Curated Meal Planning with Instacart and Walmart Cart Integration
A recipe and meal-planning engine that turns a household profile into a multi-day plan of meal cards, merges
the ingredients into one shopping list, and hands the list to an online grocery cart. Re-skin it for your
catalog, your dietitians or your customers.
1ProfileHousehold, allergens, diet style, calories
2PlanMulti-day meal plan
3Meal cardsRecipe, macros, prep and cook time
4Shopping listIngredients merged and de-duplicated
5CartInstacart or Walmart hand-off
What runs today Available now
- A proprietary library of more than 50,000 MAHA-curated recipes, searchable by meaning. The library is original NexGenHealth work: recipe frameworks drawn from open data sources, substantially transformed, and reviewed by culinary and food experts.
- Ingredients conditioned to MAHA principles, replacing items such as seed oils, margarine, refined sugar and flour, table salt, processed cheese and MSG with less-processed alternatives.
- Allergen exclusion and dietary-style filtering when recipes are selected; recipes with unknown allergen data are left out when an allergy is listed.
- Plans of several days with calories and macros per meal and weekly averages; household profiles.
- A combined weekly ingredient list with repeated items merged and pantry items optionally skipped.
- One-tap hand-off to an Instacart cart: you review, swap brands and check out on Instacart, which handles delivery.
- Kids Lunch Mode, a separate recipe pool that never mixes with adult plans.
Coming to the product
- Walmart add-to-cart option Built to order — connected with your own Walmart affiliate credentials.
- Your own retailer account: each customer connects their own Instacart or Walmart credentials Built to order.
- Recurring and scheduled delivery orders In development.
- Published MAHA rule set, a per-recipe flag, and processed-food (NOVA) classification with a retrieval filter In development.
- Condition-aware plans (sodium, glycaemic, kidney-friendly) In development.
Recipe RAGMeal cardsIngredient aggregation
Retailer hand-offAllergen filtering
“MAHA-aligned” describes the direction of our ingredient revisions, not a certification or a government endorsement.ldquo;MAHA-curated“MAHA-aligned” describes the direction of our ingredient revisions, not a certification or a government endorsement.rdquo; describes NexGenHealth's own curation of the recipe library; it is not a government certification or endorsement.
Instacart and Walmart are trademarks of their owners; NexGenHealth is not affiliated with or endorsed by either.
Delivery windows, driver tracking and checkout are provided by the retailer. Nutrition information is general and is not medical advice.
Pantry Management Android App
A household pantry app for Android: scan a barcode, see what it is, and track what is on which shelf.
In development
What the current build does
- Household pantry organised by shelf, with photos.
- Barcode scanning with product details from open food databases, and an allergen flag.
- Recipe suggestions from what is in the pantry.
- Sign-in with a NexGenHealth account; data export.
White-label and roadmap
- Rebrand and connect it to your back end for households, dietitian clients, food banks, schools, senior care or meal-kit customers Built to order.
- Expiry reminders, waste log and shopping-list sync In development.
- Public Google Play release In development.
Product data in the app includes information from Open Food Facts (Open Database Licence); images are licensed CC BY-SA.
Also in the Software Library
Components from the same platform, available separately or combined, and adaptable to your sector.
Regulated-AI Safety Layer Available now
Wraps any language model: screens each message for crisis language, neutralises prompt-injection attempts, and checks answers against the retrieved sources, deferring when support is weak.
Compliance Toolkit Available now
Field-level encryption with key rotation, an append-only consent ledger with proof of the exact text shown, and a subscription engine that enforces plans and quotas and reconciles with Stripe nightly.
Wearable Connector Hub Available now
Connect Oura, Fitbit, WHOOP and Withings accounts into one encrypted store with an access audit trail. Glucose monitors and phone health platforms are on the roadmap In development.
Health-Data Normalizers Available now
Turn the messy names on lab reports and problem lists into consistent vocabularies: 196 common blood markers with synonym and typo handling, and free-text diagnoses and ICD-10 codes mapped to a fixed condition set. Vocabularies are replaceable for your domain.
Private AI Serving Stack Available now
One local, OpenAI-compatible endpoint running open-weight models on your GPU hardware, loading and unloading models on demand, with a GPU health watchdog. We size, install and harden it.
Configurable RAG Retrieval Template Built to order
Our retrieval and orchestration pattern (per-question source selection, abstention, conflict-aware follow-up retrieval), rebuilt around your corpus, your terminology and your safety rules.
Customized for Your Sector
The same components, configured for the language, documents and rules of your field.
Clinics & digital health
Record intake, de-identified retrieval, wearable data, patient-facing plans.
Insurance & claims
Claim-document OCR, identifier scrubbing with policy and claim-number packs, grounded lookups.
Legal & compliance
Case-file OCR, redaction, and citation-backed search over your own matter documents.
HR & benefits
Employee-record scrubbing, benefits Q&A grounded in plan documents, wellness programs.
Dietitians & wellness
Client meal plans, allergen-safe recipe search, shopping lists.
Meal kits & grocers
Bring your own catalog; keep the plan, ingredient-merge and cart hand-off engine.
Schools & senior care
Menu planning pools, allergen controls, pantry tracking.
Finance & support desks
PII scrubbing for tickets, statements and transcripts before they reach analytics or models.
How You Get the Software
Self-hosted licence Available now
The software installed on your servers or in a dedicated environment, scoped and quoted to your use.
Integration and customization Available now
$349 per hour for advisory and engineering, billed to the quarter-hour. Build work is scoped and contracted separately. The first 30 minutes are free.
API access In development
Keys, usage metering and per-key limits, billed through Stripe. Tell us which components you want first and we will notify you when your place opens.
Questions Buyers Ask
Does my data leave my servers?
In our reference deployment the request path is local and no third-party AI API is called. Some libraries, however, contact outside hosts by default (for model catalogs or update checks), so a customer deployment is hardened and tested with outbound access blocked before we describe it as isolated.
Is this HIPAA compliant?
Software cannot be compliant on its own; compliance is a property of how an organization uses it. Our de-identification tool is built around the HIPAA identifier categories and publishes measured recall, and we help you plan the review your use case requires.
Can you use a model we trained ourselves?
Yes. We integrate customer models behind the same local gateway, or fine-tune adapters on your data as a scoped project. Models keep the licence terms of their upstream authors, and we tell you what those terms allow before you choose one.
Tell Us Which Software You Need
Bring your documents, your sector and your constraints. In thirty minutes we will say which components fit,
what customizing them would involve, and roughly what it would cost — before any money changes hands.
$349/hour — published, not negotiated per client
No credit card, no retainer
Response within 24 hours
Patent pending: U.S. provisional applications 63/938,050 and 63/944,649. Terminology-derived features use data
courtesy of the U.S. National Library of Medicine, National Institutes of Health, Department of Health and Human
Services; NLM is not responsible for the product and does not endorse or recommend this or any other product.
Third-party names and marks belong to their owners.