For small office cabling teams that keep estimates and change records on their NAS, Estimate Preflight brings those records into the next estimate review: what happened before, what the new draft already covers, and what still needs a conversation? It uses the selected records without requiring a separate estimating database or manual export to another AI app.
The user selects estimates and change records from two past jobs plus one current draft. Uliya's registered document expert reads those files, reports actual reading coverage, and compares supported conditions. The result is a short internal review with filenames and exact excerpts; the user decides what to ask the customer.
In one synthetic example, a past job had to be rescheduled because daytime noisy work was not permitted. The new draft says only "Work schedule to be agreed." The review asks which hours the building manager has approved. It also recognizes cable replacement already included in the draft. Wall restoration marked excluded stays excluded, while an unresolved responsibility or agreement can still prompt a question.
The five business states are included, excluded, ambiguous, not found in inspected scope, and deferred. Reading coverage is tracked separately: incomplete reading cannot establish absence. Later customer additions are not treated as proof of an estimating mistake.
The initial Skill has static checks and synthetic direct-text tests. NAS execution and extraction remain unverified. We plan to validate in Uliya's Local mode; Uliya also offers Hybrid and Cloud modes. Document processing and retention still need target validation. Testing uses synthetic or appropriately anonymized inputs, with no real confidential originals.
OpenAI ChatGPT/Codex assisted with planning, research, Skill authoring, synthetic fixtures, checks, direct-text simulations, documentation and proposal preparation. Anthropic Claude assisted with idea and submission review. The participant reviews the final materials.
Built With
- markdown
- uliya
