servers:
mcp:
transport: "http"
port: 8005
host: "127.0.0.1"
topwr_api:
host: "0.0.0.0"
port: 8000
cors_origins: "*"
rag:
max_results: 5
enable_debug: false
data_pipeline:
max_chunk_size: 30000
chunk_overlap: 200
token_limit: 65536
llm:
fast_model:
name: "gpt-5-nano"
temperature: 0.1
accurate_model:
name: "gpt-5-mini"
temperature: 0
clarin:
name: "pllum"
base_url: "https://services.clarin-pl.eu/api/v1/oapi"
gemini:
name: "gemini-2.5-flash-lite"
observability:
langfuse:
host: ${LANGFUSE_HOST}
secret_key: ${LANGFUSE_SECRET_KEY}
public_key: ${LANGFUSE_PUBLIC_KEY}
database:
name: "neo4j"
uri: ${NEO4J_URI}
username: ${NEO4J_USER}
password: ${NEO4J_PASSWORD}
nodes:
- Document
- Article
- Syllabus
- Lecture
- Course
- Professor
- Student
- Department
- Building
- Room
- Event
- Research
- Publication
- Conference
- Laboratory
- Equipment
- Project
- Grant
- Committee
- Organization
- Website
- Resource
- Schedule
- Assignment
- Grade
- Degree
- Program
- Semester
- AcademicYear
- Institution
relations:
- teaches
- enrolls_in
- attends
- conducts
- publishes
- presents
- organizes
- participates_in
- supervises
- researches
- funds
- hosts
- requires
- prerequisite_for
- belongs_to
- located_in
- schedules
- assigns
- evaluates
- awards
- manages
- collaborates_with
- cites
- references
- maintains
- develops
- supports
- coordinates
- approves
- reviews
prompts:
final_answer: |
Otrzymujesz informacje w postaci JSON w od innego LLM
z danymi pochodzącymi z bazy wiedzy.
Pytanie użytkownika: {user_input}
Informacje z bazy wiedzy (w formacie JSON):
{data}
Twoim zadaniem jest odpowiedzieć użytkownikowi na pytanie w oparciu
o te informacje - musisz wykorzystywać wszystkie dane z JSONa,
aby udzielić kompletnej odpowiedzi.
Odpowiedz w języku pytania, w sposób naturalny i zwięzły.
Jeśli nie dostaniesz informacji na temat pytania,
udziel odpowiedzi z wiedzy ogólnej na dany temat.
cypher_insert: |
Generate Neo4j Cypher statements based EXCLUSIVELY on the provided context.
Use ONLY the allowed node types and relation types.
DO NOT include any additional text or explanations.
CONTEXT: {context}
ALLOWED NODE LABELS: {nodes}
ALLOWED RELATIONSHIP TYPES: {relations}
STRICT RULES:
1. OUTPUT MUST:
- Contain ONLY executable Cypher statements
- Begin with "MERGE"
- Separate multiple statements with PIPE character (|)
- Use UNIQUE variable names (node1, node2, etc. - never reused)
- LIMIT TOKENS TO 65536 TO AVOID ERRORS WITH DEEPSEEK API
2. FOR NODES:
- MERGE each node with unique variable name
- Include 'title' and 'context' properties
- Replace Polish characters (ą→a, ć→c, ę→e, ł→l, ń→n, ó→o, ś→s, ź→z, ż→z)
- Use ONLY ASCII characters
- Escape single quotes in text with backslash (\')
3. FOR RELATIONSHIPS:
- MERGE between existing node variables
- Use ONLY allowed relationship types
- Direction matters (A→B ≠ B→A)
4. VARIABLE NAMES:
- Must be UNIQUE across entire query
- Recommended pattern: (node1), (node2), (person1), (dept1), etc.
- NEVER reuse variables - this causes errors
EXAMPLE OUTPUT:
MERGE (node1:Person {{title: 'John Smith',
context: 'Professor at UW'}})|MERGE (node2:Department
{{title: 'Computer Science', context: 'CS department'}})|MERGE
(node1)-[:works_in]->(node2)
OUTPUT MUST BE EXACTLY IN THIS FORMAT:
MERGE (...) [|MERGE (...)]* [|MERGE (...)-[:...]->(...)]*
NO OTHER TEXT OR CHARACTERS ALLOWED!
cypher_search: |
Generate ONLY valid Cypher query. No explanations.
Schema: {schema}
Question: {user_question}
Cypher:
guardrails: |
Is this about Wroclaw University of Science and Technology
(or university at all) or about another topic?
Answer ONLY: "generate_cypher" or "end"
Question: {user_question}
Answer:
