Problem
Lead applications needed to become actionable sales workflows. A score alone was not enough: it needed to connect to follow-up questions, calendar destinations and CRM records.
Context and constraints
I worked directly under the technical CEO, first as a full-stack developer and later as technical team lead. This work sat alongside backend systems, payments, HubSpot synchronization and operational automation.
Existing architecture
Lead application answers, backend services, HubSpot and scheduling destinations were parts of the workflow. The implementation used Grok, OpenAI and LangChain to support qualification.
Investigation
The useful model task was interpreting application answers to inform fit. The operational task was deciding what a supported score should do in the product and CRM. These are distinct responsibilities.
Core engineering decision
The backend owned the score-to-destination mapping. Model-assisted interpretation informed a score from 1 to 4; explicit business rules determined the next step.
Solution
Score 1 routed to a free community. Score 2 prompted more questions and could lead to a setter’s calendar. Scores 3 and 4 routed to closer calendars. I also implemented HubSpot backend updates, workflows, forms and operational tasks for failed payments.
Architecture and implementation
The integration connected application inputs, qualification, backend routing and HubSpot updates. Failed-payment recovery tasks were an adjacent operational automation, not a measured outcome of the model. The retained record does not specify exact validation schemas, retry policies or model-fallback behavior; those details are not presented as established implementation facts.
Qualification workflow
- Application answers
- OpenAI / Grok via LangChain
- Score 1–4
- Backend routing rules
Operational destinations
- Free community / follow-up / calendar
- HubSpot updates and workflows
- Operations follow-up
Result
A production qualification and routing workflow was implemented, with CRM updates and operational automation. No measured qualification uplift, time saving or revenue increase is established.
Tradeoffs
A model can assist interpretation, while business routing needs explicit ownership. For comparable systems, validation of model output, safe fallbacks, observable failures and review of qualification criteria are important production design decisions.
What I would carry forward
Retain representative inputs, expected routing and failure scenarios as an evaluation set. Measure operational outcomes before claiming that model quality or automation improved the business.