diff --git a/astroml/features/graph/snapshot.py b/astroml/features/graph/snapshot.py
index 45ec817..bc115bf 100644
--- a/astroml/features/graph/snapshot.py
+++ b/astroml/features/graph/snapshot.py
@@ -1,7 +1,8 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
-from typing import Iterable, List, Sequence, Set, Tuple
+from datetime import datetime, timedelta, timezone
+from typing import Generator, Iterable, List, Optional, Sequence, Set, Tuple
 import bisect
 
 
@@ -86,3 +87,119 @@ def snapshot_last_n_days(
     if start_ts < 0:
         start_ts = 0
     return window_snapshot(edges, start_ts, now_ts, presorted=presorted)
+
+
+# ---------------------------------------------------------------------------
+# DB-backed time-windowed snapshot slicer
+# ---------------------------------------------------------------------------
+
+@dataclass(frozen=True)
+class SnapshotWindow:
+    """A discrete time window slice ready for training."""
+    index: int          # 0-based window index (t_0, t_1, …, t_now)
+    start: datetime
+    end: datetime
+    edges: List[Edge]
+    nodes: Set[str]
+
+
+def _parse_window_size(window: str) -> timedelta:
+    """Parse a window size string like '7d', '24h', '3600s' into a timedelta."""
+    unit = window[-1].lower()
+    value = int(window[:-1])
+    if unit == "d":
+        return timedelta(days=value)
+    if unit == "h":
+        return timedelta(hours=value)
+    if unit == "s":
+        return timedelta(seconds=value)
+    raise ValueError(f"Unknown window unit '{unit}'. Use 'd', 'h', or 's'.")
+
+
+def iter_db_snapshots(
+    window: str = "7d",
+    t0: Optional[datetime] = None,
+    t_now: Optional[datetime] = None,
+    step: Optional[str] = None,
+    session=None,
+) -> Generator[SnapshotWindow, None, None]:
+    """Yield discrete time-windowed graph snapshots from the database.
+
+    Slices ``normalized_transactions`` into non-overlapping (or rolling)
+    windows from ``t0`` to ``t_now``, each of size ``window``.
+
+    Args:
+        window: Window size string, e.g. ``'7d'``, ``'24h'``, ``'3600s'``.
+        t0: Start of the first window. Defaults to the earliest timestamp in DB.
+        t_now: End of the last window. Defaults to ``datetime.now(UTC)``.
+        step: Slide step between windows (defaults to ``window`` for non-overlapping).
+              Set smaller than ``window`` for rolling windows.
+        session: SQLAlchemy session. If None, one is created via ``get_session()``.
+
+    Yields:
+        :class:`SnapshotWindow` instances in chronological order.
+    """
+    from astroml.db.schema import NormalizedTransaction
+    from sqlalchemy import select, func as sqlfunc
+
+    if session is None:
+        from astroml.db.session import get_session
+        session = get_session()
+
+    win_delta = _parse_window_size(window)
+    step_delta = _parse_window_size(step) if step else win_delta
+
+    if t_now is None:
+        t_now = datetime.now(timezone.utc)
+
+    if t0 is None:
+        result = session.execute(
+            select(sqlfunc.min(NormalizedTransaction.timestamp))
+        ).scalar()
+        if result is None:
+            return  # empty DB
+        t0 = result if result.tzinfo else result.replace(tzinfo=timezone.utc)
+
+    if t_now.tzinfo is None:
+        t_now = t_now.replace(tzinfo=timezone.utc)
+    if t0.tzinfo is None:
+        t0 = t0.replace(tzinfo=timezone.utc)
+
+    window_start = t0
+    index = 0
+
+    while window_start < t_now:
+        window_end = min(window_start + win_delta, t_now)
+
+        rows = session.execute(
+            select(
+                NormalizedTransaction.sender,
+                NormalizedTransaction.receiver,
+                NormalizedTransaction.timestamp,
+            ).where(
+                NormalizedTransaction.timestamp >= window_start,
+                NormalizedTransaction.timestamp <= window_end,
+                NormalizedTransaction.receiver.isnot(None),
+                NormalizedTransaction.sender != NormalizedTransaction.receiver,
+            ).order_by(NormalizedTransaction.timestamp)
+        ).all()
+
+        edges = [
+            Edge(src=r.sender, dst=r.receiver, timestamp=int(r.timestamp.timestamp()))
+            for r in rows
+        ]
+        nodes: Set[str] = set()
+        for e in edges:
+            nodes.add(e.src)
+            nodes.add(e.dst)
+
+        yield SnapshotWindow(
+            index=index,
+            start=window_start,
+            end=window_end,
+            edges=edges,
+            nodes=nodes,
+        )
+
+        window_start += step_delta
+        index += 1
diff --git a/astroml/features/transaction_graph.py b/astroml/features/transaction_graph.py
index b0c4f42..50048f0 100644
--- a/astroml/features/transaction_graph.py
+++ b/astroml/features/transaction_graph.py
@@ -30,7 +30,10 @@ def add_transaction(
         metadata: Optional[Dict[str, Any]] = None
     ) -> None:
         """Add a transaction edge to the graph.
-        
+
+        Self-loop edges (from_account == to_account) are silently dropped to
+        prevent infinite loops during graph traversal.
+
         Args:
             from_account: Source account identifier
             to_account: Destination account identifier
@@ -38,6 +41,8 @@ def add_transaction(
             asset: Asset type (e.g., 'USD', 'BTC', 'ETH')
             metadata: Optional transaction metadata
         """
+        if from_account == to_account:
+            return
         self.nodes.add(from_account)
         self.nodes.add(to_account)
         
diff --git a/astroml/ingestion/normalizer.py b/astroml/ingestion/normalizer.py
index e5dbda9..34b808c 100644
--- a/astroml/ingestion/normalizer.py
+++ b/astroml/ingestion/normalizer.py
@@ -5,33 +5,38 @@
 
 from astroml.db.schema import NormalizedTransaction
 from astroml.ingestion.parsers import (
+    _PATH_PAYMENT_TYPES,
     _extract_amount,
     _extract_asset,
     _extract_destination,
     _parse_datetime,
+    extract_path_payment_hops,
 )
 
 
 def normalize_operation(data: dict) -> NormalizedTransaction:
-    """Transform raw horizon operation data into a NormalizedTransaction."""
+    """Transform raw horizon operation data into a NormalizedTransaction.
+
+    For path payments use :func:`normalize_path_payment_hops` instead to
+    get one record per hop.
+    """
     op_type = data["type"]
     sender = data["source_account"]
     receiver = _extract_destination(data, op_type)
-    
+
     amount_str = _extract_amount(data)
     amount = float(amount_str) if amount_str is not None else None
-    
+
     asset_code, asset_issuer = _extract_asset(data)
-    
+
     if asset_code == "XLM" and asset_issuer is None:
         normalized_asset = "XLM"
     else:
-        # Default fallback to "UNKNOWN" if no explicit asset info is found
         normalized_asset = f"{asset_code}:{asset_issuer}" if asset_code and asset_issuer else "UNKNOWN"
 
     timestamp = _parse_datetime(data["created_at"])
     transaction_hash = data["transaction_hash"]
-    
+
     return NormalizedTransaction(
         transaction_hash=transaction_hash,
         sender=sender,
@@ -40,3 +45,32 @@ def normalize_operation(data: dict) -> NormalizedTransaction:
         amount=amount,
         timestamp=timestamp,
     )
+
+
+def normalize_path_payment_hops(data: dict) -> list[NormalizedTransaction]:
+    """Return one NormalizedTransaction per hop for a path payment operation.
+
+    Falls back to a single record (via :func:`normalize_operation`) for
+    non-path-payment types so callers can use this function uniformly.
+    """
+    if data.get("type") not in _PATH_PAYMENT_TYPES:
+        return [normalize_operation(data)]
+
+    hops = extract_path_payment_hops(data)
+    if not hops:
+        return [normalize_operation(data)]
+
+    timestamp = _parse_datetime(data["created_at"])
+    transaction_hash = data["transaction_hash"]
+
+    return [
+        NormalizedTransaction(
+            transaction_hash=f"{transaction_hash}_hop{hop['hop_index']}",
+            sender=hop["from_account"],
+            receiver=hop["to_account"],
+            asset=hop["asset"],
+            amount=hop["amount"],
+            timestamp=timestamp,
+        )
+        for hop in hops
+    ]
diff --git a/astroml/ingestion/parsers.py b/astroml/ingestion/parsers.py
index 49b6d47..e8a89c5 100644
--- a/astroml/ingestion/parsers.py
+++ b/astroml/ingestion/parsers.py
@@ -145,6 +145,11 @@ def _extract_amount(data: dict) -> Optional[str]:
         return data["amount"]
     if "starting_balance" in data:
         return data["starting_balance"]
+    # For path payments: prefer destination_amount (what receiver gets)
+    if "destination_amount" in data:
+        return data["destination_amount"]
+    if "source_amount" in data:
+        return data["source_amount"]
     return None
 
 
@@ -154,3 +159,70 @@ def _extract_asset(data: dict) -> tuple[Optional[str], Optional[str]]:
     if asset_type == "native":
         return ("XLM", None)
     return (data.get("asset_code"), data.get("asset_issuer"))
+
+
+def extract_path_payment_hops(data: dict) -> list[dict]:
+    """Decompose a path payment into ordered per-hop dicts.
+
+    Each hop dict has keys: from_account, to_account, asset_code,
+    asset_issuer, amount, hop_index, is_first_hop, is_last_hop.
+
+    Returns an empty list for non-path-payment operations.
+    """
+    if data.get("type") not in _PATH_PAYMENT_TYPES:
+        return []
+
+    sender = data["source_account"]
+    receiver = _extract_destination(data, data["type"])
+    path = data.get("path", [])  # intermediate assets
+
+    # Build asset chain: [source_asset, ...path_assets..., dest_asset]
+    def _asset_str(asset_dict: dict) -> str:
+        if asset_dict.get("asset_type") == "native":
+            return "XLM"
+        code = asset_dict.get("asset_code", "UNKNOWN")
+        issuer = asset_dict.get("asset_issuer", "")
+        return f"{code}:{issuer}" if issuer else code
+
+    src_asset_type = data.get("source_asset_type", data.get("asset_type", ""))
+    if src_asset_type == "native":
+        src_asset = "XLM"
+    else:
+        src_code = data.get("source_asset_code", data.get("asset_code", "UNKNOWN"))
+        src_issuer = data.get("source_asset_issuer", data.get("asset_issuer", ""))
+        src_asset = f"{src_code}:{src_issuer}" if src_issuer else src_code
+
+    dst_asset_type = data.get("asset_type", "")
+    if dst_asset_type == "native":
+        dst_asset = "XLM"
+    else:
+        dst_code = data.get("asset_code", "UNKNOWN")
+        dst_issuer = data.get("asset_issuer", "")
+        dst_asset = f"{dst_code}:{dst_issuer}" if dst_issuer else dst_code
+
+    path_assets = [_asset_str(p) for p in path]
+    asset_chain = [src_asset] + path_assets + [dst_asset]
+
+    # Amounts: source_amount on first hop, destination_amount on last hop,
+    # None for intermediate hops (not exposed by Horizon).
+    src_amount = data.get("source_amount")
+    dst_amount = data.get("destination_amount", data.get("amount"))
+
+    # Intermediate accounts are not exposed by Horizon; use sentinel "__path__"
+    # so the graph builder can distinguish them from real accounts.
+    n_hops = len(asset_chain) - 1
+    hops = []
+    for i in range(n_hops):
+        from_acc = sender if i == 0 else f"__path__{data['transaction_hash']}_{i}"
+        to_acc = receiver if i == n_hops - 1 else f"__path__{data['transaction_hash']}_{i + 1}"
+        amount = src_amount if i == 0 else (dst_amount if i == n_hops - 1 else None)
+        hops.append({
+            "from_account": from_acc,
+            "to_account": to_acc,
+            "asset": asset_chain[i],
+            "amount": float(amount) if amount is not None else None,
+            "hop_index": i,
+            "is_first_hop": i == 0,
+            "is_last_hop": i == n_hops - 1,
+        })
+    return hops