Add RAG basic implementation

This commit is contained in:
2026-11-09 08:43:30 +03:00
parent 447464c16c
commit 7fa1d774a0
76 changed files with 3030 additions and 124 deletions
+12 -4
View File
@@ -7,6 +7,8 @@ log_level = "INFO"
host = "0.0.0.0"
port = 8000
request_timeout_s = 120
proxy_headers = false # true behind a reverse proxy, so client addresses are real
forwarded_allow_ips = "127.0.0.1" # which proxies to trust; never "*" when the port is reachable directly
[database]
pool_size = 5
@@ -71,13 +73,19 @@ max_tasks = 60
max_task_hours = 400
max_critic_new_tasks = 10
critic_max_change_ratio = 3.0
reference_k = 5
[rag]
enabled = true
embedding_model = "text-embedding-3-small"
top_k = 5
# Retrieval of similar tasks from history. Load a corpus first: python -m estimator rag import FILE
enabled = false
provider = "openai" # openai (any OpenAI-compatible /embeddings endpoint) | hash (offline, weak)
model = "text-embedding-3-small" # the vector size is fixed at 1536 by the database schema
base_url = "https://api.openai.com/v1"
timeout_s = 60
max_retries = 2
batch_size = 64
top_k = 3 # references shown per task
min_similarity = 0.55
# api_key comes from env: EST_RAG__API_KEY (falls back to EST_LLM__OPENAI__API_KEY)
[pricing]
default_region = "ru"
+6
View File
@@ -11,6 +11,9 @@ analyst = 1500
backend = 1800
qa = 1200
devops = 1800
# no source for the next two: frontend is taken equal to backend, designer is a guess. Replace with real rates.
frontend = 1800
designer = 1500
[rates.eu]
currency = "EUR"
@@ -21,6 +24,9 @@ analyst = 75
backend = 85
qa = 60
devops = 90
# no source for the next two: frontend is taken equal to backend, designer is a guess. Replace with real rates.
frontend = 85
designer = 70
[channel_coefficient]
first = 1.0