Most clinical trial software treats the protocol as a PDF to attach somewhere and forget. We think that is a missed opportunity. The protocol is the trial; everything that gets graded, queried, or audited downstream traces back to what the protocol says.
A week about protocols, phones, and front doors
Most clinical trial software treats the protocol document as a PDF to attach somewhere and forget. We think that is a missed opportunity. The protocol is the trial; everything that gets graded, queried, or audited downstream traces back to what the protocol says.
This week we shipped the first version of a protocol-aware grading pipeline. Upload the trial protocol, and the platform parses it into structured trial metadata, embeds the document for semantic search, and threads the protocol's own evidence into every CTCAE grading run. We also overhauled the provider portal for mobile-first responsive design, stood up a passwordless demo flow at demo.burna.ai, and consolidated our seven-language translation system.
Here is what shipped, why it matters, and what is next.
Feature Highlight 1: Protocol ingestion with retrieval-augmented search
In oncology trials, the protocol is the source of truth for what counts as an adverse event, how to grade it, what the dose-modification rules are, who is eligible, and what every site is expected to report. Today, most of that information lives in a 100 to 200 page PDF that the coordinator has to translate into structured fields in an EDC and a CTMS by hand. When grading happens, the protocol's specific guidance is sitting in a binder on a shelf, not in the AI's working context.
This week we changed that. The new protocol ingestion workflow does three things:
- Parses the PDF into structured trial metadata: eligibility criteria, adverse event specifications, dose regimens, the works
- Embeds the protocol into a per-organization vector store so the entire document is semantically searchable
- Threads protocol-specific evidence into every CTCAE grading run on that trial through a knowledge-context step and a protocol-search tool available to the grading agents
When a coordinator grades an adverse event on a trial whose protocol has been ingested, the AI suggestions are no longer reasoning from generic CTCAE criteria alone. They are reasoning from the trial's actual AE-of-special-interest list, the specific dose-modification thresholds, and the protocol's grading conventions, with citations back to the page in the protocol where each rule lives.
A few practical effects:
- Setting up a new trial means uploading a protocol, not retyping it. The protocol is the source of truth, and the platform reads it the way a coordinator would.
- The extraction is reviewable before activation. A new protocol-extraction review screen in the provider app shows what the system pulled out and lets a clinician verify before going live. AI suggests, clinicians decide.
- Every grade now carries protocol citations on top of the existing CTCAE citations. Source text, CTCAE criterion, attribution rationale, and now the trial-specific protocol clause that backs the call. That is the evidence package CRO monitors are looking for.
- The same protocol-search tool is available to other workflows, so attribution and drug interaction analysis can also pull protocol context when relevant.
Why it matters: the protocol stops being a static document and becomes a live source of evidence. Coordinators spend less time configuring trials. Grades carry better citations. Monitor queries that ask "where does the protocol say this?" answer themselves.
This is the foundation for a lot of follow-on work, including protocol-aware eligibility checks at enrollment and protocol-deviation detection during the trial.
Feature Highlight 2: Provider portal goes mobile-first
Coordinators and clinicians do not always work at a desk. They triage encounters between rooms, review patients at the bedside, glance at the day's caseload from a phone in the elevator. Until this week, our provider portal was clearly designed for a wide screen and merely tolerated smaller ones.
We did a mobile-first responsive design pass across the home, encounters, and patient pages:
- Tables wrap into overflow containers with viewport-aware widths, so column counts that were unreadable on a small phone now fit cleanly inside scrollable regions
- Action buttons that crowded the header on phones collapse into a single dropdown menu on smaller viewports, preserving every action without overcrowding the layout
- Headers use responsive text sizing and stack vertically on phones, so titles do not collide with secondary metadata
- Grid spacing tightened so phones do not waste vertical space
- Top padding scales with viewport so phones do not waste a third of the screen on whitespace
Backend code touched in the same release was a formatting and logger sweep across the API surfaces, including normalized schema descriptions in protocol ingestion and standardized logging in the PRO grading action.
Why it matters: the provider portal is now usable wherever clinical work happens, not just at a workstation. This is table stakes for tablet-based bedside workflows and a real quality-of-life improvement for the rest of the team.
Feature Highlight 3: Passwordless demo mode
A persistent friction in our sales conversations: we tell people "let me show you the platform," they get excited, then they have to set up an account, get a one-time passcode in email, configure an organization, and the moment is gone.
This week we shipped demo.burna.ai. It is a passwordless demo onboarding flow gated behind an environment variable. A prospect lands on the site, fills in name, email, and phone, and is dropped into a fresh demo organization in seconds. No password, no OTP gauntlet, no manual setup. If the email already exists, the flow falls back gracefully instead of erroring out.
Underneath, we also tightened the type model and added resilient organization-slug generation with fallback logic, so two demo signups with the same name do not collide.
Why it matters: lower friction for sales conversations and partner walkthroughs. A prospect can be inside the product within 30 seconds of clicking the link.
Quality and reliability: end-to-end auth testing in CI
Two PRs landed this week that materially changed how we test authentication. A full end-to-end test suite landed first, with fixtures, page objects, global setup, and a dedicated CI workflow. The OTP-bypass route handlers run only when a dedicated test-environment variable is set, so production has no surface area.
A consolidation followed that moved all end-to-end auth bypass logic into a dedicated module behind a single registration call, removing redundant before-hook logic that was causing intermittent verification failures in preview deployments due to query/mutation timing issues. About 280 lines of test-specific code moved out of the main auth and HTTP files.
The net result: every PR now exercises sign-in and sign-up end to end on real production-shaped infrastructure, behind a single OTP bypass that no-ops with a 404 in production environments where the test variable is unset.
Improvements and bug fixes
- Translation files consolidated: each of our seven supported languages went from 22 to 30 separate JSON files into a single consolidated file with namespace keys preserved as top-level objects. Zero translation keys lost.
- Member invitation expiry hardened: invitation expiry logic now backed by a shared date utility with unit test coverage. Invite links should now consistently expire when expected.
- PRO symptom reporting fixes: resolved log emission and symptom-management edge cases.
- CORS allow-list updated: removed deprecated alpha origin, added demo.burna.ai for the new demo environment, preserved provider, beta, and staging variants.
Customer impact
Across the week's nine PRs, the headline outcomes are:
- Trial setup is now protocol-driven. Upload the protocol, and the platform reads it. Coordinators stop retyping AE specs, dose regimens, and eligibility criteria into the system.
- Every CTCAE grade can now cite the trial protocol, in addition to the standard CTCAE criterion citation. This is a real query-reduction lever for CRO and Phase 1 site monitor reviews.
- The provider portal works on phones and tablets, properly. Bedside, hallway, and between-room workflows are now first-class.
- Demos start in thirty seconds at demo.burna.ai, instead of going through the full account-setup flow.
Looking Ahead
The protocol ingestion work this week is foundational. Two near-term follow-ons we are scoping:
- Protocol-aware eligibility checks at enrollment, so we can flag patient candidates against the actual eligibility criteria from the ingested protocol
- Protocol-deviation detection during the trial, where dose adjustments, missed visits, or AE-grading patterns can be checked against the protocol's specified rules
We will also continue extending the mobile-responsive treatment to grading screens and adjudication views.
AI suggests, clinicians decide. Human-in-the-loop, always. This week the AI got better evidence to suggest from.
