File size: 10,931 Bytes
dbb04e4 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 | """
Autonomous Gap-Filling via LLM Integration (Phase 4.0)
======================================================
Bridges the GapDetector with the existing HAIMLLMIntegrator to autonomously
fill detected knowledge gaps by generating and storing synthetic memories.
Pipeline:
1. GapFiller polls GapDetector for high-priority open gaps.
2. For each gap, it constructs a prompt asking the LLM to fill it.
3. The LLM response is parsed into discrete factual statements.
4. Each statement is stored in the engine as a new memory node, tagged
with metadata: {"source": "llm_gap_fill", "gap_id": ..., "query": ...}.
5. The gap record is marked as filled in the detector registry.
Safety controls:
- Rate-limiting: max N gap-fill calls per hour (configurable).
- Confidence gate: only fill gaps that stay unresolved after min_reqs queries.
- Dry-run mode: generate responses but don't store them.
- Minimum priority threshold before triggering LLM calls.
Usage:
filler = GapFiller(engine, integrator, detector)
await filler.start() # background task
results = await filler.fill_now(n=5) # immediate fill of top-5 gaps
await filler.stop()
"""
from __future__ import annotations
import asyncio
import time
from dataclasses import dataclass, field
from datetime import datetime, timezone
from typing import Dict, List, Optional
from loguru import logger
from .gap_detector import GapDetector, GapRecord
# ------------------------------------------------------------------ #
# Configuration #
# ------------------------------------------------------------------ #
@dataclass
class GapFillerConfig:
"""Controls how/when the gap filler triggers LLM calls."""
poll_interval_seconds: float = 600.0 # check for gaps every 10 min
max_fills_per_hour: int = 20 # rate limit
min_priority_to_fill: float = 0.3 # skip low-priority gaps
min_seen_before_fill: int = 2 # gap must be seen ≥ N times
max_statements_per_gap: int = 5 # slice LLM response into pieces
dry_run: bool = False # if True: generate but don't store
store_tag: str = "llm_gap_fill" # metadata tag on stored memories
enabled: bool = True
# ------------------------------------------------------------------ #
# Prompt templates #
# ------------------------------------------------------------------ #
_FILL_PROMPT_TEMPLATE = """You are an expert knowledge assistant integrated into a cognitive memory system.
A user recently queried for information that the system could not adequately answer.
Query topic: "{query}"
Please provide a concise, factual response about this topic. Structure your answer as
{max_statements} distinct, standalone factual statements (one per line, no numbering needed).
Each statement should be directly useful for answering future questions about this topic.
Keep each statement under 150 words. Be objective and accurate.
Statements:"""
_REFINE_PROMPT_TEMPLATE = """You are helping fill a knowledge gap in a memory system.
The topic "{query}" was queried {seen} times without a satisfactory answer.
Provide {max_statements} concise factual statements that would help answer this topic.
One statement per line. Be specific, factual, and succinct (max 120 words each).
Statements:"""
# ------------------------------------------------------------------ #
# Gap filler #
# ------------------------------------------------------------------ #
class GapFiller:
"""
Autonomous LLM-driven knowledge gap filler.
Integrates with GapDetector (finds gaps) and HAIMLLMIntegrator (fills them).
"""
def __init__(
self,
engine, # HAIMEngine
llm_integrator, # HAIMLLMIntegrator
gap_detector: GapDetector,
config: Optional[GapFillerConfig] = None,
):
self.engine = engine
self.llm = llm_integrator
self.detector = gap_detector
self.cfg = config or GapFillerConfig()
self._task: Optional[asyncio.Task] = None
self._running = False
self._fill_timestamps: List[float] = [] # for rate limiting
self.stats: Dict = {
"gaps_filled": 0,
"statements_stored": 0,
"llm_calls": 0,
"errors": 0,
}
# ---- Lifecycle ----------------------------------------------- #
async def start(self) -> None:
if not self.cfg.enabled:
logger.info("GapFiller disabled by config.")
return
self._running = True
self._task = asyncio.create_task(self._poll_loop(), name="gap_filler")
logger.info(
f"GapFiller started — polling every {self.cfg.poll_interval_seconds}s"
)
async def stop(self) -> None:
self._running = False
if self._task and not self._task.done():
self._task.cancel()
try:
await self._task
except asyncio.CancelledError:
pass
logger.info("GapFiller stopped.")
# ---- Poll loop ----------------------------------------------- #
async def _poll_loop(self) -> None:
while self._running:
try:
await asyncio.sleep(self.cfg.poll_interval_seconds)
if self._running:
await self.fill_now(n=5)
except asyncio.CancelledError:
break
except Exception as exc:
logger.error(f"GapFiller poll error: {exc}", exc_info=True)
self.stats["errors"] += 1
await asyncio.sleep(60)
# ---- Fill on demand ------------------------------------------ #
async def fill_now(self, n: int = 5) -> List[Dict]:
"""
Immediately fill the top-n open gaps.
Returns:
List of fill result dicts.
"""
if not self._rate_check():
logger.warning("GapFiller rate limit reached — skipping fill cycle.")
return []
open_gaps = self.detector.get_open_gaps(top_n=n * 3) # over-fetch to filter
eligible = [
g for g in open_gaps
if g.priority_score >= self.cfg.min_priority_to_fill
and g.seen_count >= self.cfg.min_seen_before_fill
][:n]
results = []
for gap in eligible:
if not self._rate_check():
break
result = await self._fill_gap(gap)
results.append(result)
return results
# ---- Single gap fill ----------------------------------------- #
async def _fill_gap(self, gap: GapRecord) -> Dict:
"""Generate and store knowledge for a single gap."""
logger.info(
f"Filling gap '{gap.query_text[:60]}' "
f"(priority={gap.priority_score:.3f} seen={gap.seen_count})"
)
# Build prompt
prompt = _REFINE_PROMPT_TEMPLATE.format(
query=gap.query_text,
seen=gap.seen_count,
max_statements=self.cfg.max_statements_per_gap,
)
# Call LLM (runs sync _call_llm in executor)
try:
loop = asyncio.get_running_loop()
raw_response = await loop.run_in_executor(
None, self.llm._call_llm, prompt, 512
)
self._record_call()
self.stats["llm_calls"] += 1
except Exception as exc:
logger.error(f"LLM call failed for gap {gap.gap_id}: {exc}")
self.stats["errors"] += 1
return {"gap_id": gap.gap_id, "status": "error", "error": str(exc)}
# Parse into statements
statements = self._parse_statements(raw_response)
if not statements:
logger.warning(f"LLM returned no parseable statements for gap {gap.gap_id}")
return {"gap_id": gap.gap_id, "status": "empty_response"}
# Store each statement as a memory node
stored_ids = []
if not self.cfg.dry_run:
for stmt in statements:
if not stmt.strip():
continue
meta = {
"source": self.cfg.store_tag,
"gap_id": gap.gap_id,
"gap_query": gap.query_text,
"gap_signal": gap.signal,
"generated_at": datetime.now(timezone.utc).isoformat(),
"tags": ["gap_fill", "llm_generated"],
}
try:
node_id = await self.engine.store(stmt.strip(), metadata=meta)
stored_ids.append(node_id)
self.stats["statements_stored"] += 1
except Exception as exc:
logger.error(f"Failed to store gap-fill statement: {exc}")
# Mark gap as filled
self.detector.mark_filled(gap.gap_id)
self.stats["gaps_filled"] += 1
result = {
"gap_id": gap.gap_id,
"query": gap.query_text,
"status": "filled" if not self.cfg.dry_run else "dry_run",
"statements": statements,
"stored_node_ids": stored_ids,
}
logger.info(
f"Gap filled: '{gap.query_text[:50]}' "
f"→ {len(stored_ids)} statements stored"
)
return result
# ---- Helpers ------------------------------------------------- #
def _parse_statements(self, raw: str) -> List[str]:
"""
Split LLM response into individual factual statements.
Handles bullet points, numbered lists, and plain line-breaks.
"""
import re
lines = raw.strip().split("\n")
statements = []
for line in lines:
# Strip bullets / numbering
clean = re.sub(r"^[\s\-\*\d\.\)]+", "", line).strip()
if len(clean) > 20: # skip header lines / blanks
statements.append(clean)
return statements[: self.cfg.max_statements_per_gap]
def _rate_check(self) -> bool:
"""True if under the hourly rate limit."""
now = time.time()
# Keep only calls within the last hour
self._fill_timestamps = [t for t in self._fill_timestamps if now - t < 3600]
return len(self._fill_timestamps) < self.cfg.max_fills_per_hour
def _record_call(self) -> None:
"""Record a fill call timestamp for rate limiting."""
self._fill_timestamps.append(time.time())
|