Claira Health
Claira
Claira needed intake that feels supportive for patients and defensible for clinicians — automation where it saves time, human gates where it saves risk.
Automation with accountability
Phone and paper intake flooded front desks; clinicians distrusted black-box AI summaries that could misstate symptoms or medications.
Regulatory posture required clear audit trails: who saw what, what was edited, and what the patient actually attested.
The startup had to prove ROI in a single pilot site before multi-site expansion.
Structured extraction, constrained generation, explicit review
We separated structured field capture from narrative assistance — models draft within templates staff can scan in seconds.
Prompts and tools were versioned; each summary stores model ID, temperature policy, and source field hashes.
Fallback paths always exist: if LLM latency spikes, staff get the raw form — never a spinner wall.
Next.js powers the staff console; patient intake runs as a responsive flow with progress persistence and accessible error states.
LangChain-style chains handle extraction and summarization with guardrails; PII minimization redacts before model calls where possible.
Exports integrate with the clinic’s EHR-adjacent workflow via CSV and FHIR-oriented field maps (phase 1).
Deliverables
- Patient intake UX + accessibility review
- Staff review console with edit diff and sign-off
- Model governance doc + eval harness
- Audit log schema and retention policy hooks
- Training deck for front desk and nursing leads
Outcomes
- Pilot site reallocated two FTE equivalents from transcription-style work to patient-facing time.
- Staff trust scores in weekly surveys trended up after edit-in-place summaries shipped.
- Expansion LOIs referenced the review UX as the deciding factor versus a competitor “auto-send” tool.
Workflow mapping
Weeks 1–2Shadowing, risk list, field dictionary, success metrics.
Prototype
Weeks 3–4Draft flows, synthetic evals, latency benchmarks.
Pilot build
Weeks 5–7Hardened paths, logging, role-based access, training.
Go-live
Week 8Onsite support, rollback drills, hypercare window.
Client voice
“They never pitched magic. Every “AI” feature had a human checkpoint we could explain to our compliance officer.”
Related case studies.
Architecture and vendor choices were documented for the customer’s BAA posture; PHI boundaries and retention were explicit. Final compliance sign-off sits with the customer’s counsel — we delivered artifacts to support that review.
Interfaces abstract the provider; pilot used one family of models with eval snapshots so switching or fine-tuning can be staged without rewriting the app shell.
Want outcomes like Claira? Let's map your milestone.
Book a call