For medical writing and regulatory affairs

Assemble the ICH E3 CSR
without the blank page

BuildCSR assembles the ICH E3 Clinical Study Report from your TLFs, with AI-drafted section narratives from metadata only, a medical-writer review loop, and cited provenance to every supporting table.

21 CFR Part 11ICH E3 AI-drafted narrativeCited provenance Writer-approved
Use the arrows below, your keyboard's ← → keys, or swipe
The cost of the status quo

The last mile to submission is the slowest

The CSR is the final document before filing, yet every section starts from a blank page, and linking prose to the right tables is manual. The work that gates the submission has the least leverage, so the most senior writers spend their hours restating what the outputs already show.

× Blank-page drafting is slow

Writers spend hours restating what the tables already say before they can add the judgment that only a person can.

× Citations are stitched by hand

Tying each statement to its supporting TLF is manual and easy to get wrong, and a broken link is a finding waiting to happen.

× Status is hard to see

Which sections are drafted, in review, or approved lives in a tracker kept separate from the document itself.

The solution in one line

Sections as first-class units, drafted from metadata, approved by writers

BuildCSR treats each ICH E3 section as a workflow unit with its own state. An agent drafts the narrative from section metadata and TLF titles, never patient data, and the medical writer edits and approves. Every section cites its tables.

ICH E3 native

Sections carry their E3 numbers and structure, so the document is built the way reviewers expect to read it.

Drafted, then approved

The agent proposes prose, the writer refines and signs, and the human owns the text that ships.

Cited by construction

Every section links to the TLF titles that support it, captured as you write rather than reconstructed later.

Walkthrough · 1 of 7
Structure

The CSR as a tree of E3 sections

How it works
  • Sections such as 11.4.1 Demographics are first-class units, each one named and numbered the way ICH E3 lays them out.
  • Each section moves through draft, in review, and approved, and you see its state next to its title.

Why it mattersProgress is visible at the section level, and nothing ships unapproved, so the document and its status are the same thing.

app.buildcsr.io/outline
11 Efficacy Evaluation 3 sections
11.4.1 Demographics approved
narrative cited to 2 tables
11.4.2 Disposition in review
11.4.3 Exposure draft
12 Safety Evaluation 8 sections
Walkthrough · 2 of 7
Section state

Draft, in review, approved, on the record

How it works
  • Each section follows a state machine with a human gate into approved, and the gate cannot be skipped.
  • Moving a section forward writes an event, so the transition itself is part of the trail.

Why it mattersThe status of the document is unambiguous and audited, captured in the section itself rather than tracked on the side.

app.buildcsr.io/section/11.4.2
Draft In review Approved
Section 11.4.2 Disposition · current state In review · awaiting writer sign-off
Walkthrough · 3 of 7
AI drafting

A first draft from the section's metadata

How it works
  • The agent drafts from the section number, its title, and the titles of its TLFs, and it never sees patient rows.
  • The proposed prose lands in the editor as a starting point the writer can keep, rewrite, or discard.

Why it mattersWriters start from a structured draft and spend their time on judgment, not on restating what the tables already report.

app.buildcsr.io/section/11.4.1
11.4.1 Demographics narrative agent proposed
A total of 312 subjects were randomized across the two treatment arms. The treatment groups were comparable at baseline with respect to age, sex, and race, with no clinically meaningful differences observed. Baseline characteristics are summarized in the supporting tables.
Walkthrough · 4 of 7
Citations

Every statement tied to its table

How it works
  • Each narrative statement links to the supporting TLF titles, captured as the section is built.
  • The citation map travels with the section, so the link is never re-derived from memory.

Why it mattersA reviewer can confirm any claim against the exact table behind it, with no manual cross-checking of the prose against the outputs.

app.buildcsr.io/section/11.4.1/cites
StatementCited TLFLink
312 subjects randomized across two armsTable 14.1.1 Demographicslinked
Groups comparable at baselineTable 14.1.1 Demographicslinked
Disposition by treatment armTable 14.1.2 Dispositionlinked
Exposure summaryTable 14.1.3 Exposurepending
Walkthrough · 5 of 7
Review loop

The writer edits and signs

How it works
  • The writer accepts, rewrites, or rejects the draft, then approves the section, and the human owns the final text.
  • Every edit and the approval are recorded against the section as the writer works.

Why it mattersAI removes the blank page while the writer keeps authorship and accountability, so the prose that ships is the writer's own.

app.buildcsr.io/review/11.4.1
Medical writer review narrative agent proposed
A total of 312 subjects were randomized, 156 to each arm. The treatment groups were comparable at baseline with respect to age, sex, and race. The writer kept this sentence and tightened the surrounding text before approving the section.
Walkthrough · 6 of 7
Templates

Section templates you can reuse

How it works
  • Start sections from a library of E3 templates and from your own approved patterns.
  • A template seeds the structure and the expected citations, so each new study begins further along.

Why it mattersConsistent structure across studies, with less setup each time, so house style is built in rather than re-typed.

app.buildcsr.io/templates
E3 templates standard library
Safety 9 sections
Efficacy 6 sections
Disposition 3 sections
Study templates your patterns
ABC-101 approved set reusable
Walkthrough · 7 of 7
Audit

Every draft and sign-off, on the record

How it works
  • Proposals, edits, and approvals are append-only with actor, server time, and a content hash.
  • The ledger is scoped to your tenant, and updates or deletes to it are rejected even for an admin.

Why it mattersAn inspector sees how each section reached approved, and by whom, captured as the work happens rather than reconstructed under deadline.

app.buildcsr.io/audit
WhenActorEventHash
14:22:07a.duval11.4.1 approved9f2c…d1
14:10:33a.duvalwriter edit applied3b71…8e
13:58:46systemnarrative agent proposedc40a…55
13:41:12a.duvalsection opened7d18…02
Why BuildCSR

Built for the document that gates your filing

CapabilityWord + copy/pasteGeneric AI writingDocument managementBuildCSR
ICH E3 structure enforced
AI draft from metadata only, no patient data to the model
Cited provenance to supporting TLFs
Section-level state machine and sign-off
Narrative and supporting tables in one workflow
Append-only Part 11 audit trail
Built for every seat

One assembly, six points of view

Medical writer

Start from a structured draft instead of a blank page, edit and own the prose, and sign each section into approved.

Regulatory affairs

Read a document built to ICH E3 structure, with every claim cited to the table behind it before it leaves for the agency.

Lead statistical programmer

See your TLF titles flow straight into the narratives that cite them, with no re-keying of table names into prose.

Biostatistician

Confirm that each narrative statement is tied to the correct supporting output, so the science is described accurately.

Submission lead

Watch sections move to approved on one board, and ship a CSR whose status and citations are part of the document.

QA / auditor

A read-only, tamper-evident trail of every proposal, edit, and sign-off, ready for inspection on demand.

Compliance and trust

Regulatory rigor, by construction

The controls auditors look for are inherent in the system, not bolted on after the fact.

21 CFR Part 11

Gated sections with human sign-off, server-time stamping, and access control on every move into approved.

AI audit trail

Every narrative proposal is hashed and linked to the user action that accepted, edited, or rejected it.

Metadata-only AI access

The agent sees section numbers, titles, and TLF lists, never patient rows, so no patient data reaches the model.

Tenant isolation

Row-level security separates every tenant's sections and audit at the database, verified through the request path.

Assemble one of your CSR sections in a 30-day pilot

See BuildCSR draft narratives from your TLF metadata, cite every table, and move sections to approved in days. Writers own the text, and the audit trail is yours to keep.

  • We load a study's TLFs into a private workspace
  • Your writers draft, cite, and approve a set of sections
  • You keep the narratives and the citation map
  • A readout on drafting time saved
Talk to us · info@the-bdkm.com · BDKM LLC · Back to Life Sciences