Turning Work Notes into Career Stories on Amazon Bedrock
A career app turns a professional's own work notes into STAR stories, CV bullets and promotion summaries with Amazon Bedrock on Amazon Web Services (AWS), about 400 times cheaper than a career coach and in seconds instead of days.
The problem.
When a performance review, a promotion case or an interview comes up, people try to rebuild a year of work from memory. The alternative is a career coach: about NGN 8,000 (US$5.33) and 24 minutes of writing per write-up, returned days later.
Our client wanted its apps to turn a user's own notes into finished material in seconds. The notes hold employer names and private judgements, so the data had to stay private, and the model must never invent an employer, a degree or a result. As a startup, it also needed running costs that stay low.
What we built.
Three kinds of material from one note
Claude Haiku 4.5 on Amazon Bedrock, from Amazon Web Services (AWS), writes a STAR interview story, a CV bullet or a promotion summary from a user's entry. Each type has its own prompt, and output is capped at 400 tokens so answers stay short and cost stays low.
Never invent a credential
An Amazon Bedrock guardrail checks every call: it blocks requests to fabricate credentials, filters harmful content and attempts to override its instructions, and blocks or masks personal data. If the guardrail is not in place, the API refuses the request rather than calling the model.
A private data path
The apps reach an HTTP API on Amazon API Gateway through Amazon CloudFront and AWS WAF. AWS Lambda runs in private subnets of an Amazon Virtual Private Cloud (Amazon VPC) and reaches Amazon Bedrock through an interface endpoint, every request carries a signed token, and each user can only read their own entries.
Low running cost
AWS Lambda, Amazon API Gateway and Amazon Bedrock are billed per use, so cost grows with use. We operate the backend as a managed service with 25 alarms, a dashboard and a runbook.
AWS services in this engagement
The results.
A write-up now costs about US$0.0014 in Amazon Bedrock charges, and about US$0.013 fully loaded at planned volume, instead of US$5.33 for a coach, about 400 times less, and arrives in about 3 seconds instead of days. In production users generated 6,496 insights, 98.6% of generation requests completed, and the guardrail checked every one.
What we learned.
Generated code needs a checklist, not trust. An early version of the backend went out without a token check on its routes, and we caught it before real user data reached it. Every API route now verifies a signed token, and we deploy the guardrail together with the code that calls it, never separately. We also test account-level features, such as model access, before designing around them.
About Manthic.
Manthic Technologies is an AWS Advanced Tier Services Partner based in Lagos, Nigeria. We design, build and run generative AI products on AWS for our clients.
Tell us what you're trying to ship.
A 30-minute scoping call with the engineers who would do the work.