Aug to Oct 2026, owned end to end
Reference checks
Checks that start, chase and close themselves.
The problem
After an old script-and-forms setup was switched off, the ops team started every reference check by hand. They emailed every former manager, read every reply and raised every follow-up by hand.
What I built
- Checks that start themselves, with a follow-up every 24 hours until someone answers.
- An in-product form for managers, replacing Google Forms and email tables.
- A threaded inbox inside the product over AWS SES, so ops could retire a shared mailbox that kept locking them out.
- Automatic handling of no reply, wrong contact and opt-out.
- An LLM on AWS Bedrock that reads each reply, picks one of 8 labels and routes it. Unsure replies still go to a person.
- The backend and two frontends.
Stack: Node.js, BullMQ, Redis, AWS SES, AWS Bedrock, React.
What changed
- 116 to 14
- manual initiation tasks a day
- 86%
- of checks now start automatically (2,838 auto vs 818 manual)
- 22.4 to 15.4 days
- average turnaround, down 31%
- 10% to 63%
- share of tasks closed by the system
- 6.3 to 4.4
- manual actions per completed check
- 85%
- of manager responses arrive through the new form
- 16,818
- outreach emails sent by the system
Source: Production data, manual baseline 1-9 Sep vs 8-16 Oct 2026.
Your version: Replies pile up in a shared inbox?
Queued follow-ups and an LLM that reads each reply and routes it; unsure ones go to a person.
Tell me about itNext case: WhatsApp on Meta Cloud API