The product
Black Glass Leads connects lead capture, qualification, enrichment, and the work that follows. I created the product as an independent founder, owning its positioning, UX and UI, and full-stack implementation.
As of September 2026, the product is in private beta with about 10 companies. I work with these early users to refine the product before its public launch.
The central design question is simple: how do you help someone move from a collection of raw records to leads they can evaluate and act on?
My role
I work across product definition, interface design, application development, and ongoing operation. The application uses Nuxt and Vue, with TypeScript and Cloudflare infrastructure.
That breadth matters because the experience depends on the data pipeline. An interface cannot make an incomplete record trustworthy just by presenting it well. The product has to communicate what it knows and support the next useful action.
Designing around the workflow
The experience connects four related jobs:
- Capture. Bring lead information into the product through a usable entry point.
- Qualify. Evaluate how the available information fits the purpose of the workflow.
- Enrich. Add the context needed to make a record more useful.
- Act. Carry the resulting information into the team’s next step.
My design work focuses on keeping those stages understandable while accounting for the underlying processing and data-handling requirements.
Engineering the data pipeline
I built queue-based pipelines for B2B company and contact data using TypeScript, Cloudflare Workers, Queues, and D1 SQL. The ingestion API uses idempotency keys so a repeated submission can return the original response without running the scoring pipeline again. Background jobs use persisted state and retry controls to handle interrupted work.
Company records are normalized and deduplicated by domain. Contact processing uses provider identifiers, email, and LinkedIn URLs to identify duplicates. An obscured name alone is not treated as a reliable match key.
Claude integrations produce structured analysis within the enrichment workflow. Output validation and explicit failure handling keep missing or unusable results visible.
Cost is a product constraint
I built instrumentation that records provider request start and completion times, outcomes, token or credit usage, and cost information. These records support operational investigation and cost accounting. It is the same discipline I apply to my internal AI platform work.
Reconciliation controls distinguish estimates from confirmed provider costs and keep unresolved amounts explicit. Missing usage data is not treated as free usage, and recording a provider request does not itself establish a settled cost.
What this project demonstrates
Black Glass Leads brings together product judgment, design, full-stack engineering, and the ongoing responsibility of operating an independent product. It is a place where I own both the experience and the decisions behind it.