File size: 8,818 Bytes
db7ef8e | 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 | """Offline deterministic agent — works without DeepSeek keys.
Pattern-matches the question to a SQL template + UI Plan. Always produces a
valid UIPlan with citations grounded in DuckDB results.
"""
from __future__ import annotations
import re
from typing import Any
from ..retrieval.duck import run_sql
from ..ui_plan.schema import (
Citation,
Component,
Encoding,
GridCell,
Layout,
Persona,
UIPlan,
)
# ---- helpers ----------------------------------------------------------------
ACCENT = "#F47C2C"
def _kpi(id_: str, title: str, value: float | str, sub: str = "") -> Component:
return Component(
id=id_,
type="kpi",
title=title,
subtitle=sub,
data={"value": value, "format": "currency_inr" if isinstance(value, (int, float)) else "text"},
)
def _table(id_: str, title: str, rows: list[dict[str, Any]]) -> Component:
return Component(id=id_, type="data-table", title=title, data=rows[:200])
def _bar(id_: str, title: str, rows: list[dict[str, Any]],
x: str, y: str) -> Component:
return Component(
id=id_,
type="bar",
title=title,
data=rows,
encoding=Encoding(x=x, y=y, color=ACCENT),
why=f"Bar chart compares {y} across {x}.",
)
def _line(id_: str, title: str, rows: list[dict[str, Any]],
x: str, y: str) -> Component:
return Component(
id=id_,
type="line",
title=title,
data=rows,
encoding=Encoding(x=x, y=y, color=ACCENT),
why=f"Line chart traces {y} over {x}.",
)
def _wrap(question: str, narrative: str, components: list[Component],
source_table: str, sql: str) -> UIPlan:
grid: list[GridCell] = []
y = 0
for i, c in enumerate(components):
if c.type == "kpi":
grid.append(GridCell(id=f"g{i}", component=c.id, w=4, h=2, x=(i % 3) * 4, y=0))
else:
grid.append(GridCell(id=f"g{i}", component=c.id, w=12, h=5, x=0, y=y or 2))
y += 5
return UIPlan(
persona=Persona(name="analyst", accent=ACCENT, mood="investigative"),
layout=Layout(grid=grid),
components=components,
narrative=narrative,
citations=[
Citation(
id="cite1",
source=source_table,
snippet=sql[:280],
)
],
)
# ---- intent matchers --------------------------------------------------------
def _try_top_funders(q: str) -> UIPlan | None:
if not re.search(r"\b(top|biggest|largest)\b.*\bfunder", q, re.I):
return None
sql = """
SELECT funder_name,
ROUND(SUM(actual_total)::DOUBLE, 2) AS actual_inr
FROM bva
WHERE funder_name IS NOT NULL AND funder_name <> 'NONE'
GROUP BY funder_name
ORDER BY actual_inr DESC NULLS LAST
LIMIT 10
"""
res = run_sql(sql)
if res.error or not res.rows:
return None
top = res.rows[0]
grand = sum(float(r["actual_inr"] or 0) for r in res.rows)
comps = [
_kpi("k1", "Top funder",
top["funder_name"], f"₹{top['actual_inr']:,.0f}"),
_kpi("k2", "Top-10 funders total", grand, f"FY26 actual"),
_kpi("k3", "Funders shown", len(res.rows), "rows"),
_bar("c1", "Top 10 funders by FY26 actual spend",
res.rows, x="funder_name", y="actual_inr"),
_table("t1", "Detail", res.rows),
]
narrative = (
f"Across the top 10 funders, **{top['funder_name']}** leads at "
f"₹{float(top['actual_inr']):,.0f}. Combined top-10 actual is "
f"₹{grand:,.0f}, sourced from the `bva` ledger."
)
return _wrap(q, narrative, comps, "bva", sql)
def _try_monthly_burn(q: str) -> UIPlan | None:
if not re.search(r"\b(monthly|month|trend|burn|over time)\b", q, re.I):
return None
cols = [f"actual_{m}" for m in [
"01042025", "01052025", "01062025", "01072025", "01082025", "01092025",
"01102025", "01112025", "01122025", "01012026", "01022026", "01032026"]]
months = ["Apr-25", "May-25", "Jun-25", "Jul-25", "Aug-25", "Sep-25",
"Oct-25", "Nov-25", "Dec-25", "Jan-26", "Feb-26", "Mar-26"]
union = " UNION ALL ".join(
f"SELECT '{m}' AS month, {i:02d} AS ord, "
f"ROUND(SUM({c})::DOUBLE,2) AS actual_inr FROM bva"
for i, (m, c) in enumerate(zip(months, cols), 1)
)
sql = f"SELECT * FROM ({union}) ORDER BY ord"
res = run_sql(sql)
if res.error or not res.rows:
return None
total = sum(float(r["actual_inr"] or 0) for r in res.rows)
peak = max(res.rows, key=lambda r: float(r["actual_inr"] or 0))
comps = [
_kpi("k1", "FY26 actual (Apr–Mar)", total, "from bva monthly cols"),
_kpi("k2", "Peak month", peak["month"], f"₹{peak['actual_inr']:,.0f}"),
_kpi("k3", "Months", len(res.rows), "covered"),
_line("c1", "Monthly burn — FY26",
res.rows, x="month", y="actual_inr"),
_table("t1", "Monthly detail", res.rows),
]
narrative = (
f"The FY26 monthly burn totals ₹{total:,.0f}, peaking in "
f"**{peak['month']}** at ₹{float(peak['actual_inr']):,.0f}. "
f"Trend is computed from the per-month `actual_*` columns of `bva`."
)
return _wrap(q, narrative, comps, "bva", sql)
def _try_top_categories(q: str) -> UIPlan | None:
if not re.search(r"\bcategor", q, re.I):
return None
sql = """
SELECT category_name,
ROUND(SUM(actual_total)::DOUBLE, 2) AS actual_inr
FROM bva
WHERE category_name IS NOT NULL
GROUP BY category_name
ORDER BY actual_inr DESC NULLS LAST
LIMIT 15
"""
res = run_sql(sql)
if res.error or not res.rows:
return None
comps = [
_kpi("k1", "Top category", res.rows[0]["category_name"],
f"₹{res.rows[0]['actual_inr']:,.0f}"),
_bar("c1", "Top 15 spend categories", res.rows,
x="category_name", y="actual_inr"),
_table("t1", "Detail", res.rows),
]
return _wrap(
q,
f"Top spending category is **{res.rows[0]['category_name']}** "
f"at ₹{float(res.rows[0]['actual_inr']):,.0f}.",
comps, "bva", sql)
def _try_open_po(q: str) -> UIPlan | None:
if not re.search(r"\b(open\s*po|purchase order|po\b|vendor)\b", q, re.I):
return None
sql = "SELECT * FROM open_po LIMIT 200"
res = run_sql(sql)
if res.error or not res.rows:
return None
comps = [
_kpi("k1", "Open POs", len(res.rows), "rows"),
_table("t1", "Open Purchase Orders", res.rows),
]
return _wrap(q, f"{len(res.rows)} open purchase orders found.",
comps, "open_po", sql)
def _try_overview(q: str) -> UIPlan:
"""Always-on default overview."""
funders = run_sql("SELECT COUNT(DISTINCT funder_name) AS n FROM bva "
"WHERE funder_name IS NOT NULL AND funder_name <> 'NONE'")
projects = run_sql("SELECT COUNT(DISTINCT project_id) AS n FROM bva")
grand = run_sql("SELECT ROUND(SUM(actual_total)::DOUBLE,2) AS v FROM bva")
region = run_sql("""
SELECT COALESCE(region,'(unspecified)') AS region,
ROUND(SUM(actual_total)::DOUBLE,2) AS actual_inr
FROM bva
GROUP BY region
ORDER BY actual_inr DESC NULLS LAST
LIMIT 12
""")
fcount = funders.rows[0]["n"] if funders.rows else 0
pcount = projects.rows[0]["n"] if projects.rows else 0
gv = grand.rows[0]["v"] if grand.rows else 0
comps = [
_kpi("k1", "Distinct funders", fcount, "in BVA ledger"),
_kpi("k2", "Distinct projects", pcount, "tracked"),
_kpi("k3", "FY26 actual (gross)", gv, "₹"),
_bar("c1", "Spend by region", region.rows,
x="region", y="actual_inr"),
_table("t1", "Region breakdown", region.rows),
]
return _wrap(q,
f"BVA spans **{fcount} funders** and **{pcount} projects** with "
f"a gross actual of ₹{float(gv):,.0f} for FY26.",
comps, "bva", region.sql)
MATCHERS = [
_try_top_funders,
_try_monthly_burn,
_try_top_categories,
_try_open_po,
]
def offline_plan(question: str) -> UIPlan:
for fn in MATCHERS:
try:
plan = fn(question)
except Exception: # noqa: BLE001
plan = None
if plan is not None:
return plan
return _try_overview(question)
|