lucaborella89
commented
3 weeks ago Sequence 1) Redline needs to be broken; user prompt is going to hit the graph (great for profiling CDM object) --> LLM --> CDM object as output (Ian to be confirmed). The graph and its architecture is needed for a) user experience (show me how you model the output of a credit swap); b) synthetic data maybe at a later stage.
Sequence 2) change the output of the LLM to be CDM legal agreement data, the work will potentially show how CSA map into CDM, how AI can be used to generate CDM (code) from CSAs
Sequence 3) will share an update soon
Sequence 4) will share an update soon
Meeting Minute
We recapped what was discussed in last week (technical) call and discussed the proposed flow of the OSFF Workshop (1st of October). It is worth using a modified-ORR diagram to illustrate all sessions together. 1) Generate Synthetic CDM Data -> Provectus 2) Map Credit Support Annexes with CDM -> ISDA 3) Find Granular Obligations -> AI4FINANCE 4) Generate Synthetic DRR Data (TBC) -> TradeHeader
We discussed how the 4 sessions will be demoed during the OSFF Workshop. Given the time constrains, each presenter (Provectus, ISDA, AI4FINANCE, TradeHeader) will use its own resources (i.e. login to own instances) to showcase / demo at OSFF.
We discussed the need to formalise the contribution processes around LLM Exploration Project. For the timebeing a quick fix is keep using this FINOS Labs repo till we don't have the regular GitHub at the level of other projects/sigs
We discussed the messaging for the workshop: "this is an example of what can be done with LLMs, for more use cases refer to the pre-OSFF London workshop https://www.finos.org/blog/finos-llm-exploration-use-case-identification"