Deployment and Upgrade of Marketing Content for Joint-Stock

What the data for this category looks like

Joint-stock bank marketing content data mainly comes from the internal marketing material library, CRM customer management system, and retail wealth management business system. The data update rhythm adjusts with marketing campaign cycles. Regular brand activities are updated quarterly. Temporary promotions or hot marketing can trigger updates at any time. Documents are divided into three categories by type: graphic promotional materials, customer-group tailored scripts, and activity rule descriptions. Each document includes fields such as title, core content, applicable customer group tags, delivery channel, and validity period. Validity period is measured in days. Customer group tags are stored as comma-separated strings.

What constraints these characteristics impose on deployment and upgrade

Multi-source data access requires configuring multiple interface authentication rules during deployment to avoid permission conflicts during cross-system synchronization. Non-fixed-frequency updates require supporting hot loading of the material library during the upgrade phase. The latest marketing content can be synchronized without restarting the service. Multi-type document structures and dedicated fields require configuring differentiated parsing rules during deployment. This adapts to field extraction logic for different materials. Marketing content compliance requirements require embedding a sensitive word verification module during the deployment phase. This ensures generated or recalled content meets regulatory requirements. Sudden update scenarios require ensuring service availability during the upgrade process. This avoids affecting normal marketing activity calls due to deployment operations.

How to set the configurations

Configuration Item	Recommended Value	Rationale
`UPLOAD_FILE_MAX_SIZE`	`200 MB`	Most joint-stock bank marketing materials are posters or long copy. This value covers most single-file scenarios
`PARSE_FILE_TIMEOUT_SECONDS`	`300 seconds`	Multi-page graphic or long-text marketing materials require sufficient parsing time to avoid mid-process timeout interruptions
`maxContext`	`800–1200 characters`	The core information of marketing script materials falls within this length range. This filters redundant content and improves recall efficiency
`Recall count`	`Top 8 entries`	Balances customer group precise matching and content richness. Avoids excessive recall results increasing subsequent processing time
`Similarity threshold`	Calibrated based on actual testing	Adjust based on customer group tag matching degree to filter low-match non-target marketing materials
`Rerank result count`	`Top 3 entries`	Marketing content display needs to be concise and intuitive. This value ensures core information reaches users first

The parameter values provided on this page are common recommendations used as a starting point for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require specific analysis. It is recommended to test on your own samples before finalizing the configuration.

Three common mistakes

Phenomenon: one API login reports Username or password error. Cause: The API_KEY environment variable was not correctly mapped during deployment, causing the authentication logic to fail to read valid credentials.
Phenomenon: Error response from daemon is returned when running docker compose up -d. Cause: The local image was not fully pulled or the port is occupied, and port occupancy and image integrity were not checked in advance.
Phenomenon: Knowledge base search takes too long. Cause: The Recall count configuration is set too high, and vector database index optimization is not enabled, resulting in an overly large retrieval scope and excessive processing load.

How to confirm the configuration is correct

Upload a typical marketing script document. Check if the background parsing result extracts dedicated fields such as applicable customer group tags and delivery channels to confirm that the parsing rules take effect.
Run the docker compose ps command. Confirm that all containers are in the Up state, with no abnormal exits or restart records.
Initiate a marketing content recall test. Check if the number of returned results conforms to the configured recall rules to verify that the parameter configuration takes effect.
Temporarily adjust the value of the similarity threshold. Initiate a test to confirm that low-match results are filtered or retained, verifying that the threshold configuration takes effect.

Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.