How the system
actually thinks.
Grounded. Auditable. Built so cost and policy detail are never hallucinated.
Plain-language financial planning that touches clinical decisions is a regulatory minefield, and most LLMs hallucinate on cost and policy detail. Nestie’s stack is built so neither happens — and so clinics, partners and families can see exactly how it works.
RAG · Multi-model · Guardrails · Structured outputs
Approach
Built to be honest about what AI can — and cannot — do.
Four design principles shape every layer of the stack. They are boring on purpose: trust at this stage of a family’s life is not earned with novelty.
Grounded in retrieval, never memory.
Every numeric and clinical claim is anchored to a retrieved source snippet from a curated UK/EU corpus. Nothing is invented from model weights.
Educational. Never advisory.
Generation is restricted to planning and education. Clinical-advice requests are routed to clear disclaimers — never answered by the model.
A neutral planner, not a salesperson.
Non-clinical, non-lender. Clinics, insurers, banks and families can align on a single high-intent funnel without conflicts of interest baked in.
Free for families. Monetised with partners.
No signup, no paywall. Trust is earned at the family layer; regulated financial products are matched through partners once a clinic is live.
The pillars
Eight pieces. One single-pass plan.
Retrieval-augmented generation
Grounded in a continuously refreshed UK/EU corpus — clinic pricing, treatment protocols, medication regimens, insurance policy text and cross-border regulatory rules.
Anchored to sources, not memory
Dense embeddings + sparse keyword + reranking. Every claim is anchored to a retrieved source snippet — nothing invented from model memory.
The right model for the right job
Long-context LLMs draft full plans in one pass; lightweight models route conversation; specialised numeric models decompose cost via chain-of-thought reasoning.
Structured cost-forecasting
Totals decomposed by cycle, medications, scans and contingency events. Line-item explainable — not a black-box estimate a patient cannot interrogate.
Coverage-gap extraction
Clause-level extraction over insurance policy text — surfacing what is covered, partially covered, or excluded before the financial decision is made.
Clinic and pathway matching
Vector similarity across cost, success rate, specialism, location and language — so families and clinics meet on the most relevant pathway.
Regional, not translated
Modelled natively per region — so a UK patient and a patient considering Turkey, Spain or Greece both see jurisdiction-specific guidance.
Educational by design
Input and output guardrails keep generation inside educational and planning territory. Clinical-advice requests are routed to clear disclaimers, never answered.
The flow
From a question to a plan in one pass.
Most assistants either improvise or stall. Nestie composes the answer once, with sources attached, and writes a plan the family keeps coming back to.
- 01Ask
Family question, in their words.
"What does IVF actually cost me in London?" Free-text, multi-language, no signup required.
- 02Route
Intent and jurisdiction read.
A lightweight router determines region, scope, and which retrieval indexes to query — UK schedules vs cross-border options.
- 03Retrieve
Hybrid retrieval, reranked.
Dense embeddings + sparse keywords + reranker pull only the relevant clinical and financial source snippets from the curated corpus.
- 04Reason
Multi-model composition.
A long-context LLM drafts the plan in one pass. A specialist numeric model decomposes cost by cycle, medications, scans and contingency.
- 05Structure
Plain language out, JSON underneath.
The family reads a plain-language plan. Partners and dashboards consume the same plan as structured, auditable JSON.
- 06Match
Context-aware partner handoff.
When the family is ready to act, the platform tells you what type of financing or partner to look for — without initiating any transaction or making a referral.
< 8s
Median plan generation, end-to-end
100%
Numeric claims tied to retrieved sources
100%
Clinical-advice requests routed to disclaimer
A real answer
Every number is cited.
The family sees a plain-language plan. Behind every figure sits a retrieved snippet from a clinic schedule, an insurance policy or a published protocol — not the model’s memory.
A single IVF cycle in London typically costs £8,000–£15,000 per cycle1. With medications, scans and a likely second cycle, families plan for £18,000–£32,0002.
If you have a private insurance plan, IVF itself is usually excluded3, but diagnostics and consultations are sometimes covered — I can check yours line-by-line.
Want me to break this down by clinic, or factor in your insurance?
- 1PRICING
Clinic schedule
IVF cycle fee, London partner clinic
Retrieved · v2026-Q1 · 0.94
- 2CLINICAL
Protocol
Median IVF cycles to live birth, UK 35–39 cohort
Retrieved · published 2025 · 0.89
- 3POLICY
Insurance policy
Fertility exclusions clause, private PMI plan
Clause-level NLP · 0.91
Reranker score shown alongside each retrieval. If no source clears the threshold, the model declines — it does not improvise.
Guardrails
The boundary, drawn in plain sight.
Clinical advice belongs to clinicians. Nestie explains the system around them — cost, coverage, pathways — and hands off everything else, with intent and context preserved.
We answer
Educational and planning
- Explain how an IVF cycle is priced
- Compare cost ranges across clinics
- Break down what insurance covers vs excludes
- Plan financing options across cycles
- Surface jurisdiction differences (UK / Spain / Turkey / Greece)
We never answer
Clinical advice, diagnosis, prescription
- Recommend whether to do a fresh or frozen transfer
- Interpret an AMH or hormone test result
- Choose between donor and own-egg pathways
- Diagnose a fertility condition
- Prescribe medication doses
Refusal is not silence. Clinical-advice requests are routed to a clear disclaimer and, where the partner clinic supports it, a direct handoff to a coordinator.
The output
Plain language for the family. Structured JSON for everyone else.
The same plan is rendered as a portable document for the family, and as an auditable JSON payload for clinic dashboards, lenders, insurers and payment providers.
{
"patient_id": "anon_4f9c",
"region": "en-GB",
"treatment": "ivf",
"cycles_planned": 2,
"cost_breakdown": [
{ "item": "Cycle 1 base fee", "low": 8000, "high": 15000, "src": 1 },
{ "item": "Medications", "low": 1200, "high": 2500, "src": 1 },
{ "item": "Scans + bloods", "low": 600, "high": 1100, "src": 1 },
{ "item": "Cycle 2 (likely)", "low": 8000, "high": 13000, "src": 2 }
],
"totals": { "low": 17800, "high": 31600, "currency": "GBP" },
"coverage": {
"ivf": { "status": "excluded", "src": 3 },
"scans": { "status": "partial", "src": 3 },
"consult": { "status": "covered", "src": 3 }
},
"financing_type": "personal_loan_recommended",
"guardrails": { "clinical_advice": "refused" }
}Portable schema
A documented plan schema with cycles, line items, coverage gaps, financing options and partner matches — versioned and stable.
Family-side persistence
Plans are saved and reopenable as treatment progresses. No signup. Free at the family layer, always.
Partner handoff, in context
When a family is ready to act, the JSON travels with them — so the lender or coordinator does not start from zero.
See it in motion
Architecture is easier to trust when you can watch it run.
Book a 30-minute walkthrough — we’ll trace a real family plan end-to-end, surface every retrieval and reranker score, and show you the JSON your dashboard would consume.
Best Femtech Solution of the Year