Keep the voice.
Eleven brand fields carry identity, audience, personality, phrases, values, style, and example writing into generation and repurposing.
Projects / ContentForge AI
Generative AI + content workflows
ContentForge AI
Your voice, defined once.
Carried through every draft.
- Generative AI
- Brand voice
- Next.js
- TypeScript
- Prisma / PostgreSQL
- Clerk
- Google Calendar
Define a brand voice → build a brief → generate content → keep and plan the work.
00 / The idea
More than
another prompt box.
Marketing teams and independent creators need consistent voice across different formats. ContentForge AI makes that context reusable: define the brand, write a brief, compare different creative approaches, and carry the result into a searchable library and planning calendar.
- Challenge
- Generic copy, repeated prompting, and drafts scattered across disconnected tools.
- Approach
- Reusable brand context + format-aware generation + a durable content workflow.
- Workspace
- Individual authenticated workspaces with ownership-scoped records.
- Boundary
- Planning and Google Calendar synchronization, not automatic social publishing.
01 / Product walkthrough
From a brand voice to a planned release.
01/6
The product promise
Create marketing content that still sounds like your brand.
Your workspace
Usage, saved content, brands, and upcoming plans in one dashboard.
Define the voice
Identity, audience, values, writing style, and examples become reusable context.
Write the brief
Choose a format, topic, audience, tone, and up to three distinct variations.
Keep the work
Search, edit, repurpose, duplicate, and schedule saved content.
Plan the next release
A monthly planning calendar, with optional Google Calendar synchronization.
02 / System architecture
A complete workflow.
Clear boundaries.
The Next.js application owns both the product interface and API routes. Clerk establishes identity, Prisma scopes persisted records, and a server-only adapter connects generation to the selected AI provider.
One application. Explicit server boundaries.ContentForge AI / System map
ProductAuthenticated APIServices & state
Next.js interfacedashboard · create · library · calendar · brands
Clerk sessionsign in → authenticated workspace
Typed API clientvalidated requests · loading & error states
Route handlersresolve active user · validate input
Ownership & limitsuser-scoped queries · atomic AI allowance
Application servicesgenerate · repurpose · schedule · synchronize
AI provider adapterOpenAI-compatible / Anthropic / mock
Prisma → PostgreSQLbrands · content · schedules · usage
Google Calendaroptional consent-based event synchronization
Change the approach.
Direct, Story-driven, and Educational strategies produce different structures rather than simply requesting paraphrases.
Keep the result.
Saved drafts remain editable, searchable, reusable, and schedulable. A connected calendar is an optional extension.
03 / The generation pipeline
One brief.
Different directions.
Format constraints, audience, objective, tone, length, keywords, and CTA are combined with the brand profile. Up to three independent requests run in parallel; their JSON title and body are validated before previewing or saving.
- X / Twitter
- Marketing email
- Ad copy
- Product description
- Blog post
A shared voice. Distinct creative directions.ContentForge AI / System map
ContextStrategy fan-outValidate & retain
Brand voiceidentity · audience · phrases · style · examples
Content briefformat · topic · objective · tone · length · CTA
Prompt builderformat rules + voice + chosen strategy
Directclear benefit · immediate hook · focused CTA
Story-drivennarrative opening · tension · resolution
Educationalexplanation · practical steps · takeaway
Provider responsesbounded retries · structured title / body
JSON validationparse response · reject unreadable output
Preview or savetransactional draft records when requested
Provider configuration
Keep the workflow.
Switch the model.
- OpenAI-compatible
- gpt-4oDefault model identifier; the model and base URL can be configured for a compatible gateway.
- Anthropic
- claude-sonnet-4-20250514Configurable default for the Messages API adapter.
- Mock
- Deterministic templatesStrategy-aware local outputs for offline development and tests; not model inference.
These are configurable code defaults, not claims about which model ran every demo. Brand instructions guide generation; human review is still needed for facts, voice, and compliance.
04 / Reuse & planning
A draft shouldn't
be a dead end.
Repurpose a saved piece into another format while retaining the voice and source relationship. Schedule it on the monthly calendar, then optionally synchronize a Google event. Local plans remain saved when external synchronization fails.
From a saved idea to a planned releaseContentForge AI / System map
LibraryAdaptPlanOptional sync
Saved contentsearch · edit · copy · duplicate
Repurposesource text + brand voice → another format
Scheduleplatform + date/time → monthly calendar
Google eventopt-in synchronization · retryable errors
Schedule mutations use database transactions. Google updates use a per-schedule advisory lock and deterministic event identity.
A workspace anchored to one identityContentForge AI / System map
IdentityCreative recordsOperational records
UserClerk identity · active workspace
Lifecycle synchronizationsigned webhooks · active-user checks
Brandvoice fields · owned by user
Contentformat · body · status · brand · source link
ScheduledContentdate/time · platform · event sync state
AiUsageunique user + UTC day · operation count
GoogleCalendarConnectionconsent preference · external account ID
05 / Engineering underneath
Useful AI needs
reliable plumbing.
Authentication, ownership checks, bounded retries, atomic usage reservations, and transactional scheduling support the creative workflow. The implementation includes Vitest coverage for these boundaries; this case study does not claim newly measured benchmark results.
- 01
auth
Resolves the Clerk identity to an active database user and provisions the workspace before authenticated rendering.
- 02
prompt-builder
Combines the brief, brand voice, format constraints, and variation strategy into a structured generation prompt.
- 03
generators
Runs independent variation requests in parallel and adapts saved content into another format.
- 04
provider
Switches OpenAI-compatible, Anthropic, or mock adapters; bounds retries and handles upstream timeouts.
- 05
validation / limits
Validates request fields and generated JSON, caps brands, and atomically reserves daily AI operations.
- 06
scheduling / calendar
Keeps content and planning records consistent, then optionally synchronizes Google Calendar events.
Explicit allowances.
Up to three brands and 50 generation or repurposing operations per user per UTC day. Accepted operations consume quota even if the provider later fails.
Retry selectively.
Transient network errors, rate limits, and selected server failures receive bounded retries. Unreadable responses produce a recoverable user-facing error.
Save first. Sync safely.
Database transactions protect planning records. Per-schedule advisory locks and stable Google event IDs reduce concurrent-update duplication.