Lead enrichment pipeline
A six-phase run where something always fails partway
A long-running enrichment and scoring pipeline that survives the third-party rate limits and timeouts it is built on.
- Year
- 2026
- phases, resumable
- 6
- Stack
- Python, MongoDB, LLM scoring
- Status
- Not public
The problem
Build a qualified prospect list: pull company records in paginated batches, enrich them, then score each against an ideal-customer profile.
The constraint
The engineering problem is not the scoring. It is that a six-phase run across rate-limited third-party APIs takes long enough that something will fail partway through, and restarting from the beginning wastes both hours and paid API calls.
The decision
Every phase checkpoints its state to disk. Supervisor scripts watch the process, notice when it has died or wedged, and resume it from the last checkpoint rather than the start.
A pipeline that cannot resume is a pipeline you run once, and one you run once is one you cannot iterate on.
At a glance
- Phases
- 6, each checkpointed
- Recovery
- Automatic resume
- Scoring
- LLM against an ICP