altinn / ai Goto Github PK
View Code? Open in Web Editor NEWGenerative AI exploration
License: MIT License
Generative AI exploration
License: MIT License
Before attempting to fix specific quality issues related to the documentation training dataset, we should document them and perform a cost-benefit analysis together with the relevant teams.
There are several hundred Altinn II services (infopath forms) that need to be converted to Altinn 3 apps, prior to the phase-out of Altinn II.
A certain proportion of these require few or no changes (unchanged XSD data model), while others will need some "refurbishing". The hypothesis is that a lot of time can be saved if one can import the existing form as a starting point for further work.
The tool https://github.com/Altinn/altinn2-convert converts XSN files to Altinn 3 apps and handles data model, GUI (pages and elements), texts, bindings between data model and GUI, etc. For example, the Financial Supervisory Authority has tried converting all its approximately 120 services.
Additional Information
Expected challenges with importing from Infopath files (.CAB, zipped XSN files):
Expected challenges with importing from PDF:
An overview of relevant training datasets for RAG-style prompt generation and chaining:
In a multi-query workflow, a single incoming query can be forwarded to multiple search backends and combined using techniques such as Reciprocal rank fusion
See also Ensemble Retriever
Pagesense support similarity and hybrid search within the same index, see Pagesense docs
Tried hosting blog articles on the Github wiki space included in our subscription, but it doesn't provide basic functionality such as ordering by blog post date or displaying posts in a navigation component.
Easiest option is to use a default blog theme for a static site generator such as Astro.
No response
Propose a practical approach to improving the current documentation
Criteria:
Suggestions:
No response
No response
No response
We have demands for delivering complete documentation in both norsk bokmål/nynorsk and english (DPG). Translating is time consuming for all teams.
Generate translated pages on commit and just mark them "AI translated" and/or let teams review the translations before published.
In my opinion this should be prioritized, and i could probablly be based on work alerady done with the assistant? Please advice? @altinnadmin @bdb-dd
Verify that our repositories are correctly configured to meet our unique file naming conventions.
No response
No response
No response
Identify and setup Azure resources needed to run Assistant-style AI apps for experimentation (not full production requirements)
Although initial costs are expected to be low due to intermittent use and no requirement for dedicated GPU resources, we need to develop a good understanding for how costs accrue for different approaches.
Potential dependency on #8
Name is just a suggestion, sure if "ai" is already used to indicate "application insights".
"assistant" could also be used to group several AI apps, without the "ai" designation.
Central to the definition of an Altinn 3 app are three well defined JSON schemas. JSON files conforming to these schemas are used to define data models, layouts and text translations.
Studio is a user friendly editing interface for these files, which are saved in a Git repository.
Our hypothesis is that Studio Assistant could be added to Studio with minimal integration effort and significant functional augmentation. Specifically, Studio Assistant can deliver functionality that would normally require significant effort to design a usable visual interface for.
Make a good list of useful prompts. Here are some ideas to get started:
"Flytt adressefeltene til å være i en egen gruppe som heter 'Postadresse'"
"Endre de merkerte felt til å være påkrevd"
"Dupliser gruppenummer 2"
"Hvilke felt er påkrevd?"
"Gå til neste felt som mangler oversettelse til nynorsk"
No response
No response
No response
Ref:
Query relaxation and scoping
https://www.algolia.com/blog/ux/query-relaxation-and-scoping-as-part-of-semantic-search/
Small to big retrieval
https://www.youtube.com/watch?v=ihSiRrOUwmg
Include a reference to this nice introduction video, explaining RAG concepts:
https://www.youtube.com/watch?v=T-D1OfcDW1M
There are many examples of existing PDF and HTML forms across the Norwegian public service.
We should experiment and evaluate existing tools combined with new LLM-based techniques for interpreting existing form definitions and generating equivalent Altinn 3 apps.
Not ready for a full scale conversion effort until initial evaluation has been reviewed.
No response
No response
In order to gain interoperability between different platforms and solutions, we need good metadata about the data used in a service. This job with data classification and populating the data catalogues on data.norge.no has been lagging behind for years.
Could a simple AI application feed on raw data an achieve a >80% coverage on a dataset in the data-catalogue that could raise coverage and data quality in order to stimulate to data-driven services in A3, and with a more runtime approach to the data-registries at data.norge.
As a starter this could be done with data managerd/owned by digdir that is not yet described in FDK. If the case is valid, this approach could be used in mapping and compiling other metadata-classes that we dont have general overview on like for instance processing of personal data (could be scraped from the "personvernærklæring"), in order to create cool personal data management applications.
Reports that may be relevant in the regard of this case:
https://www.regjeringen.no/contentassets/0e36c85fcfe143a5b626c53cf292cb3b/altinns-innspill-vedlegg-1---konseptet-digitale-meg-2019.07.05.pdf
https://www.digdir.no/datadeling/innsynslosning-tekniske-og-juridiske-muligheter/3465
https://www.digdir.no/digital-samhandling/konseptskisse-realisering-av-en-innbyggerorientert-digital-assistent/2949
In order to facilitate safe and secure experimentation with machine learning practices, we need to prioritize defining initial guidelines and restrictions for training datasets, agent functionality scope, deployment and testing.
No response
No response
No response
No response
No response
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.