Given a financial identifier, the algo tries to identify the type of identifier, and a place to get the data.

Within finance there are so many identifiers: legal entities, instrument identifiers, product codes, exchnge names,

The consumer has an identifier represented as a string. They don't know what it is. For example, someone sees an ID in a text file, or on a screen. Rather than search multple systems to find the identifier, they can simply paste it in. This isn't a 'Google' for financial data, because that's Google. The limitation of google is that many data are licensed so not publically available. Additionally, the results from Google often don't lead to machine reable results without large engineering costs.

String: identifier

Can be multiple results:

String: identifier Enum: IdentifierType (BBG ID, SEC LEI, MIC Code, ISO code, etc) MatchType: Exact | Pattern | Fuzzy | Guess/Regex DataSource: Link to where the data can be sourced from

• Exact = An exact match based on a brute force lookup, or lookup based on a Elasticache mechanism
• Patter = Using a neural network, the algo has determined a likely identifier type
• Fuzzy = This isn't an exact match, but is very simliar to a record that exists (computationally expensive!)
• Guess/Regex = Hard-coded/regex based matching

The results would be offered in a freemium model. E.g. the first X queries per day are without cost; or perhaps only the first result is returned. This would allow us to gauge product/market fit from a basic level.

If the algorithm is succesful then add-ons can be numerous to expand revenue possibilities:

• Email add-ons; search the email for identifiers, and make them click-able (e.g. highlight a bond isin and provide a link to the refernece data, or price source)
• Powering internal search: allow organizations to use their own systems for lookup, so that the algo returns company internal links.
• Get more challenging. E.g. interpret descriptions such as US 10 6/10 -> Us 10 yr bond paying .6%; then find the identifiers

• The algo would be a key part; presuming a CNN of some kind
• The service could be operated as a lambda function in AWS
• Any caching could be done with DynamoDB
• Exposed to users using a public swagger endpoint via AWS API gateway
• The model would be hosted on an S3 bucket

Fixed costs likely to be less than $10/month.

BBG821804393 -> could use a regex patern or CNN to recognize the first 3 digits are BBG and match on that basis ISIN -> US9183893; again could use regex SEC LEI -> 7ZW8QJWVPR4P1J1KQY45 -> how to recognize? https://www.gleif.org/lei/7ZW8QJWVPR4P1J1KQY45

Day 1:

Use a CNN to quickly guess the ID types. Then use regex to exclude impossible matches. Then lookup 3rd party services to affirm likelihood of being a real identifier.

Day 2:

Multi-word pattern matching; e.g. "UST10 Comdty" -> https://www.openfigi.com/id/BBG0036CPY08

Day 3:

More identifier types, such as legal entity identifiers, and ability to map free text to idenfitiers. Eg. Volkswagon AG to VW AG.

• Scrapers to get more source data, if partners cannot provide large data sets
• Find free places to send people to after match (e.g. exchanges websites)

Day 4:

Ability to show relationships such as VWAGY ADR is a child of VW AG listed in Germany

See Planning/Search/README.md