I am using Ollama v0.2.3 on Windows with tinyllama, locally installed, and langchain4j v0.32.0.
I followed a very simple example of sending a chat query to Ollama.
To my surprise I got back a very random response, having nothing to do with my query. After some debugging it turned out that the Java code was sending the JSON query to /api/chat with "role":"USER", but it only works with "role":"user". In other words:

curl http://localhost:11434/api/chat -d "{"model":"tinyllama","messages":[{"role":"user","content":"Provide three short bullet points explaining why Java is awesome"}],"options":{},"stream":false}"

This works fine.

curl http://localhost:11434/api/chat -d "{"model":"tinyllama","messages":[{"role":"USER","content":"Provide three short bullet points explaining why Java is awesome"}],"options":{},"stream":false}"

This gives a random response.

The issue appears to be in the JSON serialization which takes the Role enum, which is all uppercase.

Since I am only toying around I replaced that enum class with a patched version which has additional serialization instructions to make the role values lowercase:

package dev.langchain4j.model.ollama;

import com.google.gson.annotations.SerializedName;

enum Role {
  @SerializedName("system") // added to make it work
  SYSTEM,
  @SerializedName("user") // added to make it work
  USER,
  @SerializedName("assistant") // added to make it work
  ASSISTANT;
}

Overall, however, serializing into uppercase for the role seems to be wrong. Should Ollama deal with e.g. USER (instead of user), or should langchain4j send the role value in lowercase? Is this a bug in Ollama/langchain4j, or am I missing some configuration?