# RASA core to interact with two different NLU models based on condition

**URL:** <https://forum.rasa.com/t/rasa-core-to-interact-with-two-different-nlu-models-based-on-condition/17886>\
**Category:** Rasa Open Source\
**Created:** [October 2, 2019, 9:18pm UTC](https://forum.rasa.com/t/rasa-core-to-interact-with-two-different-nlu-models-based-on-condition/17886 "2019-10-02T21:18:15Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![abhi\_bh\_nlp](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/abhi_bh_nlp/32/5905_2.png) [@abhi\_bh\_nlp](https://forum.rasa.com/u/abhi_bh_nlp)\
**Post date:** [October 2, 2019, 9:18pm UTC](https://forum.rasa.com/t/rasa-core-to-interact-with-two-different-nlu-models-based-on-condition/17886/1 "2019-10-02T21:18:15Z")

</div>

Hi, I have trained two different NLU models say model 1 and model 2, you can think of it as a hierarchical architecture where in i wish to call the NLU model2 only if the intent predicted from the 1st model is say “A” else will continue using model1. I want to have this architecture when RASA CORE interacts with NLU. How do we achieve this.

 ![Capture1](https://europe1.discourse-cdn.com/flex013/uploads/rasa/original/2X/0/05168da8982eabea527b9868b672d7a2991bb0de.png)

---

<div class="post-metadata">

**Author:** ![Ghostvv](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/ghostvv/32/211_2.png) [@Ghostvv](https://forum.rasa.com/u/Ghostvv)\
**Post date:** [October 7, 2019, 8:56am UTC](https://forum.rasa.com/t/rasa-core-to-interact-with-two-different-nlu-models-based-on-condition/17886/2 "2019-10-07T08:56:01Z")

</div>

You’d need to implement custom nlu pipeline (at least for prediction) for this and a custom action in core to call another nlu model and to forget previous utterance and overwrite it with the new one. You can take a look how we implemented `Supervised Response Selector`: docs: [Responses](https://rasa.com/docs/rasa/core/responses/#custom-nlg-service), blog: [Integrate response retrieval models in assistants built with Rasa](https://blog.rasa.com/response-retrieval-models/)