Prosjekt 2026

[ Vis hele beskrivelsen ]

Telebuster is a prototype dialog system from year 2000, that understands conversations about travel information and can provide schedule information. The task will be to make it ready for 2026, by updating all the sub-systems to the latest technology, and by connecting it to GTFS, and recent large language models (LLMs) to make it available with updated schedules, anywhere in the world, all the time.

The existing system is built in layers with Prolog - PHP - and Java. It would be an advantage to reduce the pipeline by using only two layers. In light of the new LLMs, it would be interesting to explore how difficult/easy it is to get the 2000  dialog model up and running again (with the help of LLMs), vs implementing the same (or better) functionality with “modern approaches”. In terms of Sustainability, it should also be documented how much power is needed to run the old system vs the new systems.

Fluent knowledge of Norwegian and English is a huge advantage, but clever use of Google translate etc. can compensate for lack in experience.

[ Skjul beskrivelse ]

[ Vis hele beskrivelsen ]

In Trondheim we have the bus oracle (BusTUC) which has answered tens of millions English and Norwegian questions about the buses in Trondheim since 1997.

You can try it yourself at http://busstuc2.idi.ntnu.no

On the other hand, there are also several services that answer many of the same questions, just by selecting location, destination and time constraints.

The goal is to investigate the best way for different groups of users, in different situations, to get travel information. The work can follow up the FUIROS project, investigating Speech, GPS, Maps, or Real-time data integration and route updates.

One example problem that could be investigated is that when people ask the Oracle and get an answer they are not happy with, they have to start from scratch again, because the system "forgets everything" between questions. How can we make the Oracle react appropriately to each single user's needs, by remembering everything that they have asked before? How does that fit within GDPR regulations?

Some existing work to build on can be found here.

[ Skjul beskrivelse ]

[ Vis hele beskrivelsen ]

For years HR software have tried to tackle this problem by using manual labor. However, the problem lies in accessibility and the fact that is easier to ask the question directly and get a qualified answer to your problem on the fly, does not make it a sustainable approach.

Meanwhile, GPT (ChatGPT and other Generative Pretrained Transformers) is increasingly able to create consistant answers to general answers in realtime, both in English, Norwegian and other languages.

The current task is to asses the possibility of using GPT also on sensitive data. The project can be to automate our manual assistant (going onto maternity leave now) and make a "Mia-prototype" for actual use at NTNU.

IDEAS:

Lag et grensesnitt for uthenting av snittlønn for enkeltmedlemmer.

• +/- 5 år fra egen alder. 
• Basert på eget (oppgitt) ansattnr eller Innlogging med FEIDE
• Blokker etter gjentatte ulovlige innlogginger
• Andre interessante statistikker: 2.5.3-fordeling på avtaler
• etc
• Studie: Datasikkerhet, GDPR, sikker/anonym innlogging.
• SIKT-godkjenning må søkes, og avklaring med ledelsen på NTNU om data-tilgang (NDA-avtale) er en forutsetning for dette prosjektet.

[ Skjul beskrivelse ]