MOI parses email threads plus PDF and Excel attachments, extracts RFQ fields, then supports code mapping, human validation, and SQL Server ingestion.
The client is an aviation parts and MRO (maintenance, repair and overhaul) provider supplying airlines and operators, handling a high volume of parts quotations and technical specification matching.
Requests for quotation almost never arrive structured. They sit in email bodies, in the context of a long reply chain, and in PDF and Excel attachments whose layout differs with every customer.
A single RFQ can carry a hundred or more line items, each needing around ten key fields extracted: part number, quantity, condition (new, overhauled, serviceable) and more. Staff re-keyed them line by line — slow, and error-prone in a domain where a mis-keyed part number has airworthiness and safety consequences rather than merely commercial ones.
On top of that, the material descriptions customers use do not match the company's internal aviation parts master data. The same part number can be expressed completely differently by different customers, so a mapping step is required before anything can enter the business system.
MOI starts from the email itself, processing the EML together with its PDF and Excel attachments, and extracts RFQ information into structured fields.
Parsing combines rules and a large model: the parts that are regular and positionally fixed are handled deterministically by rules; the parts whose layout varies and require reading context go to the model. The division avoids both failure modes — the instability of leaving everything to a model, and the brittleness of rules that break the first time a new customer template appears.
The ten extracted business fields then go through code-table mapping, aligning customer terminology to the aviation parts master. Tables of more than a hundred rows are processed in batches, so extraction quality does not degrade with an over-long context.
Mapped results are checked by a person before being written transactionally into the SQL Server business system — parsing extracts and proposes, but a human owns the commit.
The path from email to business system is connected end to end. Staff no longer key in hundred-row material lists by hand, and RFQ response time drops materially.
Code-table mapping normalizes customer terminology to the parts master, so the material basis for a quote stays consistent across customers and across the people handling them.
With RFQ data structured and stored, supplier comparison and historical price benchmarking have a usable data foundation instead of depending on searching back through email.
Airworthiness and compliance requirements are surfaced during field validation, so problems appear at quotation time rather than at delivery.