flare / chat_handler.py
ciyidogan's picture
Update chat_handler.py
e72db51 verified
raw
history blame
7.62 kB
"""
Flare – Chat Handler (v1.3 Β· robust trimming + param validation)
=================================================================
"""
import re, json, uuid, sys, httpx, commentjson
from datetime import datetime
from typing import Dict, List, Optional
from fastapi import APIRouter, HTTPException, Header
from pydantic import BaseModel
from commentjson import JSONLibraryException
from prompt_builder import build_intent_prompt, build_parameter_prompt, log
# ───────────────────────── HELPERS ───────────────────────── #
def _trim_response(raw: str) -> str:
"""
Remove everything after the first logical assistant block or intent tag.
Also strips trailing 'assistant' artifacts and prompt injections.
"""
# Stop at our own rules if model leaked them
for stop in ["#DETECTED_INTENT", "⚠️", "\nassistant", "assistant\n", "assistant"]:
idx = raw.find(stop)
if idx != -1:
raw = raw[:idx]
# Normalise selamlama
raw = re.sub(r"Hoş[\s-]?geldin(iz)?", "Hoş geldiniz", raw, flags=re.IGNORECASE)
return raw.strip()
def _safe_intent_parse(raw: str) -> (str, str):
"""Extract intent name and extra tail."""
m = re.search(r"#DETECTED_INTENT:\s*([A-Za-z0-9_-]+)", raw)
if not m:
return "", raw
name = m.group(1)
tail = raw[m.end():]
return name, tail
# ───────────────────────── CONFIG ───────────────────────── #
CFG = commentjson.load(open("service_config.jsonc", encoding="utf-8"))
PROJECTS = {p["name"]: p for p in CFG["projects"]}
APIS = {a["name"]: a for a in CFG["apis"]}
SPARK_URL = CFG["config"]["spark_endpoint"].rstrip("/") + "/generate"
ALLOWED_INTENTS = {"flight-booking", "flight-info", "booking-cancel"}
# ───────────────────────── SESSION ───────────────────────── #
class Session:
def __init__(self, project: str):
self.id = str(uuid.uuid4())
self.project = PROJECTS[project]
self.history: List[Dict[str, str]] = []
self.vars: Dict[str, str] = {}
self.awaiting: Optional[Dict] = None
log(f"πŸ†• Session {self.id}")
SESSIONS: Dict[str, Session] = {}
# ───────────────────────── SPARK ───────────────────────── #
async def spark_generate(s: Session, prompt: str, user_msg: str) -> str:
payload = {
"project_name": s.project["name"],
"user_input": user_msg,
"context": s.history[-10:],
"system_prompt": prompt
}
async with httpx.AsyncClient(timeout=60) as c:
r = await c.post(SPARK_URL + "/generate", json=payload)
r.raise_for_status()
d = r.json()
raw = (d.get("assistant") or d.get("model_answer") or d.get("text", "")).strip()
log(f"πŸͺ„ Spark raw: {raw[:120]!r}")
return raw
# ───────────────────────── FASTAPI ───────────────────────── #
router = APIRouter()
@router.get("/") # health
def health(): return {"ok": True}
class Start(BaseModel): project_name: str
class Body(BaseModel): user_input: str
class Resp(BaseModel): session_id: str; answer: str
@router.post("/start_session", response_model=Resp)
async def start_session(req: Start):
if req.project_name not in PROJECTS:
raise HTTPException(404, "project")
s = Session(req.project_name)
SESSIONS[s.id] = s
greet = "Hoş geldiniz! Size nasıl yardımcı olabilirim?"
return Resp(session_id=s.id, answer=greet)
@router.post("/chat", response_model=Resp)
async def chat(body: Body, x_session_id: str = Header(...)):
if x_session_id not in SESSIONS:
raise HTTPException(404, "session")
s = SESSIONS[x_session_id]
user = body.user_input.strip()
s.history.append({"role": "user", "content": user})
# ---------- follow-up ----------
if s.awaiting:
ans = await _followup(s, user)
s.history.append({"role": "assistant", "content": ans})
return Resp(session_id=s.id, answer=ans)
# ---------- intent detect ----------
p = build_intent_prompt(
s.project["versions"][0]["general_prompt"],
s.history, user, s.project["versions"][0]["intents"])
raw = await spark_generate(s, p, user)
if raw == "":
fallback = "Üzgünüm, mesajınızı anlayamadım. Lütfen tekrar dener misiniz?"
s.history.append({"role": "assistant", "content": fallback})
return Resp(session_id=s.id, answer=fallback)
# small-talk yolu
if not raw.startswith("#DETECTED_INTENT"):
clean = _trim_response(raw)
s.history.append({"role": "assistant", "content": clean})
return Resp(session_id=s.id, answer=clean)
intent, tail = _safe_intent_parse(raw)
if intent not in ALLOWED_INTENTS or len(user.split()) < 3:
clean = _trim_response(tail)
s.history.append({"role": "assistant", "content": clean})
return Resp(session_id=s.id, answer=clean)
cfg = _find_intent(s, intent)
if not cfg:
err = "Üzgünüm, anlayamadım."
s.history.append({"role": "assistant", "content": err})
return Resp(session_id=s.id, answer=err)
answer = await _handle_intent(s, cfg, user)
s.history.append({"role": "assistant", "content": answer})
return Resp(session_id=s.id, answer=answer)
# ────────────────── INTENT / PARAM / API HELPERS ────────────────── #
def _find_intent(s, name): return next((i for i in s.project["versions"][0]["intents"] if i["name"] == name), None)
def _missing(s, cfg): return [p["name"] for p in cfg["parameters"] if p["variable_name"] not in s.vars]
async def _handle_intent(s, cfg, user):
missing = _missing(s, cfg)
if missing:
prmpt = build_parameter_prompt(cfg, missing, user, s.history)
raw = await spark_generate(s, prmpt, user)
if raw.startswith("#PARAMETERS:") and not _proc_params(s, cfg, raw):
missing = _missing(s, cfg)
if missing:
s.awaiting = {"intent": cfg, "missing": missing}
cap = next(p for p in cfg["parameters"] if p["name"] == missing[0])["caption"]
return f"{cap} nedir?"
s.awaiting = None
return await _call_api(s, cfg)
async def _followup(s, user):
cfg = s.awaiting["intent"]; miss = s.awaiting["missing"]
prmpt = build_parameter_prompt(cfg, miss, user, s.history)
raw = await spark_generate(s, prmpt, user)
if not raw.startswith("#PARAMETERS:") or _proc_params(s, cfg, raw):
return "Üzgünüm, anlayamadım."
miss = _missing(s, cfg)
if miss:
s.awaiting["missing"] = miss
cap = next(p for p in cfg["parameters"] if p["name"] == miss[0])["caption"]
return f"{cap} nedir?"
s.awaiting = None
return await _call_api(s, cfg)
def _proc_params(s, cfg, raw):
try: d = json.loads(raw[len("#PARAMETERS:"):])
except: return True
for pr in d.get("extracted", []):
pcfg = next(p for p in cfg["parameters"] if p["name"] == pr["name"])
if not re.fullmatch(pcfg.get("validation_regex", ".*"), pr["value"]):
return True
s.vars[pcfg["variable_name"]] = pr["value"]
return Fals