-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
1101 lines (962 loc) · 33.5 KB
/
index.html
File metadata and controls
1101 lines (962 loc) · 33.5 KB
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
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Synthetic Data for Reliable AI Systems</title>
<style>
* {
margin: 0;
padding: 0;
box-sizing: border-box;
}
:root {
--primary: #2563eb;
--primary-light: #dbeafe;
--text: #1e293b;
--text-muted: #64748b;
--bg: #f8fafc;
--white: #ffffff;
--accent: #0891b2;
--warning: #d97706;
--success: #059669;
}
body {
font-family: 'Segoe UI', system-ui, -apple-system, sans-serif;
background: var(--bg);
color: var(--text);
line-height: 1.6;
}
/* Slide container */
.slide {
display: none;
min-height: 100vh;
padding: 60px 80px;
background: var(--white);
position: relative;
}
.slide.active {
display: flex;
flex-direction: column;
}
/* Slide header */
.slide-header {
margin-bottom: 40px;
}
.slide-number {
font-size: 14px;
color: var(--text-muted);
margin-bottom: 8px;
font-weight: 500;
}
h1 {
font-size: 48px;
font-weight: 700;
color: var(--primary);
line-height: 1.2;
}
h2 {
font-size: 36px;
font-weight: 600;
color: var(--text);
line-height: 1.3;
}
h3 {
font-size: 24px;
font-weight: 600;
color: var(--accent);
margin-top: 32px;
margin-bottom: 16px;
}
/* Content */
.slide-content {
flex: 1;
display: flex;
flex-direction: column;
justify-content: center;
}
.slide-content.top {
justify-content: flex-start;
}
p {
font-size: 24px;
margin-bottom: 20px;
color: var(--text);
}
.subtitle {
font-size: 28px;
color: var(--text-muted);
margin-top: 16px;
}
ul {
list-style: none;
padding-left: 0;
}
li {
font-size: 24px;
margin-bottom: 16px;
padding-left: 36px;
position: relative;
}
li::before {
content: "→";
position: absolute;
left: 0;
color: var(--primary);
font-weight: bold;
}
li.check::before {
content: "✓";
color: var(--success);
}
li.warn::before {
content: "✗";
color: var(--warning);
}
li.bullet::before {
content: "•";
}
/* Highlight boxes */
.highlight {
background: var(--primary-light);
border-left: 4px solid var(--primary);
padding: 24px 32px;
margin: 24px 0;
font-size: 24px;
}
.highlight.insight {
background: #ecfdf5;
border-left-color: var(--success);
}
.highlight.warning {
background: #fef3c7;
border-left-color: var(--warning);
}
/* Big quote / statement */
.big-statement {
font-size: 42px;
font-weight: 600;
color: var(--text);
text-align: center;
padding: 40px;
line-height: 1.4;
}
.big-statement em {
color: var(--primary);
font-style: normal;
}
/* Two column layout */
.two-col {
display: grid;
grid-template-columns: 1fr 1fr;
gap: 60px;
margin-top: 20px;
}
.col h3:first-child {
margin-top: 0;
}
/* Stats */
.stats {
display: flex;
justify-content: space-around;
margin: 40px 0;
}
.stat {
text-align: center;
}
.stat-number {
font-size: 64px;
font-weight: 700;
color: var(--primary);
line-height: 1;
}
.stat-label {
font-size: 18px;
color: var(--text-muted);
margin-top: 8px;
}
/* Steps */
.steps {
counter-reset: step;
}
.step {
display: flex;
align-items: flex-start;
margin-bottom: 24px;
}
.step-num {
background: var(--primary);
color: white;
width: 36px;
height: 36px;
border-radius: 50%;
display: flex;
align-items: center;
justify-content: center;
font-weight: 600;
font-size: 18px;
margin-right: 20px;
flex-shrink: 0;
}
.step-content {
flex: 1;
}
.step-title {
font-size: 22px;
font-weight: 600;
color: var(--text);
margin-bottom: 4px;
}
.step-desc {
font-size: 18px;
color: var(--text-muted);
}
/* Code/example */
.example {
background: #f1f5f9;
border-radius: 8px;
padding: 20px 24px;
font-family: 'Consolas', 'Monaco', monospace;
font-size: 18px;
margin: 16px 0;
overflow-x: auto;
}
/* Navigation */
.nav-hint {
position: fixed;
bottom: 20px;
right: 20px;
font-size: 14px;
color: var(--text-muted);
background: var(--white);
padding: 8px 16px;
border-radius: 4px;
box-shadow: 0 2px 8px rgba(0,0,0,0.1);
}
/* Progress bar */
.progress {
position: fixed;
top: 0;
left: 0;
height: 4px;
background: var(--primary);
transition: width 0.3s ease;
}
/* Title slide special */
.title-slide {
justify-content: center;
text-align: center;
}
.title-slide h1 {
font-size: 56px;
margin-bottom: 24px;
}
.title-slide .subtitle {
font-size: 32px;
}
.presenter {
margin-top: 60px;
font-size: 22px;
color: var(--text);
}
.presenter-name {
font-weight: 600;
font-size: 26px;
margin-bottom: 8px;
}
.presenter-title {
color: var(--text-muted);
margin-bottom: 4px;
}
.presenter-link {
color: var(--primary);
text-decoration: none;
}
.presenter-link:hover {
text-decoration: underline;
}
/* Section divider */
.section-slide {
justify-content: center;
align-items: center;
background: linear-gradient(135deg, var(--primary) 0%, var(--accent) 100%);
}
.section-slide h2 {
color: white;
font-size: 48px;
text-align: center;
}
.section-slide .section-num {
color: rgba(255,255,255,0.7);
font-size: 24px;
margin-bottom: 16px;
}
/* Print styles */
@media print {
@page {
size: A4 landscape;
margin: 0;
}
body {
background: white;
}
.slide {
display: flex !important;
page-break-after: always;
page-break-inside: avoid;
width: 297mm;
height: 210mm;
min-height: 210mm;
max-height: 210mm;
padding: 15mm 20mm;
overflow: hidden;
background: var(--white);
}
.slide:last-child {
page-break-after: auto;
}
.nav-hint, .progress {
display: none !important;
}
.slide-header {
margin-bottom: 8mm;
}
.slide-content {
justify-content: flex-start;
}
.slide-content.top {
justify-content: flex-start;
}
h1 {
font-size: 32pt;
margin-bottom: 4mm;
}
h2 {
font-size: 24pt;
}
h3 {
font-size: 14pt;
margin-top: 6mm;
margin-bottom: 3mm;
}
p {
font-size: 14pt;
margin-bottom: 4mm;
}
li {
font-size: 14pt;
margin-bottom: 3mm;
padding-left: 8mm;
}
.big-statement {
font-size: 22pt;
padding: 8mm;
}
.stat-number {
font-size: 36pt;
}
.stat-label {
font-size: 11pt;
}
.stats {
margin: 8mm 0;
}
.highlight {
font-size: 13pt;
padding: 4mm 6mm;
margin: 4mm 0;
}
.highlight ul {
margin-top: 3mm;
}
.example {
font-size: 11pt;
padding: 4mm 5mm;
margin: 3mm 0;
}
.two-col {
gap: 10mm;
margin-top: 4mm;
}
.step {
margin-bottom: 5mm;
}
.step-num {
width: 8mm;
height: 8mm;
font-size: 12pt;
margin-right: 4mm;
}
.step-title {
font-size: 14pt;
margin-bottom: 1mm;
}
.step-desc {
font-size: 12pt;
}
.slide-number {
font-size: 10pt;
margin-bottom: 2mm;
}
.title-slide h1 {
font-size: 36pt;
margin-bottom: 5mm;
}
.title-slide .subtitle {
font-size: 20pt;
}
.presenter {
margin-top: 15mm;
font-size: 14pt;
}
.presenter-name {
font-size: 16pt;
margin-bottom: 2mm;
}
.section-slide h2 {
font-size: 32pt;
}
.section-slide .section-num {
font-size: 16pt;
}
/* Force backgrounds to print */
.section-slide {
-webkit-print-color-adjust: exact;
print-color-adjust: exact;
}
.highlight, .highlight.insight, .highlight.warning {
-webkit-print-color-adjust: exact;
print-color-adjust: exact;
}
.step-num {
-webkit-print-color-adjust: exact;
print-color-adjust: exact;
}
}
</style>
</head>
<body>
<div class="progress" id="progress"></div>
<!-- Slide 1: Title -->
<div class="slide title-slide active">
<h1>Synthetic Data for Reliable AI Systems</h1>
<p class="subtitle">Case Study: Text-to-SQL for Business Users</p>
<div class="presenter">
<div class="presenter-name">Dima Kuchin</div>
<div class="presenter-title">Fractional AI CTO</div>
<a href="https://linkedin.com/in/kuchin" class="presenter-link">linkedin.com/in/kuchin</a>
</div>
</div>
<!-- Slide 2: The Promise vs Reality -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">The Problem</div>
<h2>The Promise vs. The Reality</h2>
</div>
<div class="slide-content">
<div class="stats">
<div class="stat">
<div class="stat-number">90%</div>
<div class="stat-label">Accuracy on benchmarks</div>
</div>
<div class="stat">
<div class="stat-number">→</div>
<div class="stat-label"></div>
</div>
<div class="stat">
<div class="stat-number">30-50%</div>
<div class="stat-label">Accuracy on real enterprise data</div>
</div>
</div>
<div class="big-statement">
Works great in demos.<br>Fails in production.
</div>
<div class="highlight insight">
<strong>Thesis:</strong> Synthetic data is what bridges demo to production.
</div>
</div>
</div>
<!-- Slide 3: What Text-to-SQL Must Do -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">The Problem</div>
<h2>What Text-to-SQL Must Actually Do</h2>
</div>
<div class="slide-content top">
<p>Translate <strong>business intent</strong> into correct SQL.</p>
<div class="example">
"Show me our best customers"
</div>
<ul style="margin-top: 24px;">
<li>What does "best" mean? Revenue? Orders? Tenure? Margin?</li>
<li>What's a "customer"? Active only? Include churned?</li>
<li>What time period? All time? This year? Trailing 12 months?</li>
</ul>
<div class="highlight warning" style="margin-top: 32px;">
Every company defines these differently. Generic models guess <b>every single time</b>.
</div>
</div>
</div>
<!-- Slide 4: Why It Fails -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">The Problem</div>
<h2>Why It Fails in Practice</h2>
</div>
<div class="slide-content top">
<h3>The Semantic Gap</h3>
<p>Business language ≠ Database structure</p>
<h3>The Plausibility Trap</h3>
<p>SQL runs successfully, looks correct, <em>answers the wrong question</em></p>
<h3>Company-Specific Everything</h3>
<ul>
<li class="bullet">Jargon: "QBR", "MQL", "Land and Expand"</li>
<li class="bullet">Fiscal calendars: FY starts in February</li>
<li class="bullet">Metric definitions: "Active user" means 5 different things</li>
</ul>
</div>
</div>
<!-- Slide 5: The Trust Problem -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">The Problem</div>
<h2>The Trust Problem</h2>
</div>
<div class="slide-content">
<div class="big-statement">
80% accuracy is <em>worse than useless</em>
</div>
<p style="text-align: center; margin-top: 40px;">
Users can't tell which 20% is wrong.<br>
They either distrust everything or trust everything.
</p>
<div class="highlight" style="margin-top: 40px;">
<strong>Requirement:</strong> Transparency, verification, graceful failure.
</div>
</div>
</div>
<!-- Slide 6: Section - Solution -->
<div class="slide section-slide">
<div class="section-num">Part 2</div>
<h2>What a Working Solution Requires</h2>
</div>
<!-- Slide 7: Building Blocks -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">The Solution</div>
<h2>Building Blocks</h2>
</div>
<div class="slide-content top">
<div class="two-col">
<div class="col">
<h3>Semantic Layer</h3>
<p>Business terms → SQL mappings</p>
<h3>Verified Query Bank</h3>
<p>Known-good question→SQL pairs</p>
<h3>Ambiguity Handling</h3>
<p>Ask clarifying questions when uncertain</p>
</div>
<div class="col">
<h3>Feedback Loop</h3>
<p>Learn from corrections over time</p>
<h3>Evaluation Framework</h3>
<p>Company-specific benchmarks</p>
<h3>Trust Indicators</h3>
<p>Show reasoning, flag uncertainty</p>
</div>
</div>
</div>
</div>
<!-- Slide 8: The Bootstrap Problem -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">The Solution</div>
<h2>The Bootstrap Problem</h2>
</div>
<div class="slide-content top">
<p>To build this system, you need:</p>
<ul style="margin: 24px 0;">
<li>Hundreds of verified question→SQL pairs</li>
<li>Company-specific knowledge encoded</li>
<li>Edge cases and variations covered</li>
</ul>
<div class="highlight warning">
<strong>Manual creation:</strong><br>
10-30 min per query × 300 queries = <strong>weeks of analyst time</strong>
</div>
<p style="margin-top: 32px; font-size: 28px; text-align: center;">
You need training data to build the system.<br>
Building training data manually doesn't scale.
</p>
</div>
</div>
<!-- Slide 9: Section - Why Synthetic -->
<div class="slide section-slide">
<div class="section-num">Part 3</div>
<h2>Why Synthetic Data</h2>
</div>
<!-- Slide 10: The Data Paradox -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">Why Synthetic Data</div>
<h2>The Data Paradox</h2>
</div>
<div class="slide-content">
<div class="big-statement">
To generate realistic queries,<br>you need realistic data.
</div>
<div class="big-statement" style="margin-top: 20px;">
Real data is exactly what<br><em>you cannot use</em>.
</div>
<ul style="margin-top: 32px;">
<li class="warn">PII and confidential information</li>
<li class="warn">Compliance and governance policies</li>
<!-- <li class="warn">Even on-premise LLMs face audit barriers</li> -->
</ul>
</div>
</div>
<!-- Slide 11: What You CAN Use -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">Why Synthetic Data</div>
<h2>What You CAN Use</h2>
</div>
<div class="slide-content top">
<div class="two-col">
<div class="col">
<h3>Structure</h3>
<ul>
<li class="check">Database schema (DDL)</li>
<li class="check">Column metadata & comments</li>
<li class="check">Semantic layer definitions</li>
</ul>
</div>
<div class="col">
<h3>Usage</h3>
<ul>
<li class="check">Dashboard definitions (SQL, LookML, ...)</li>
<li class="check">Query log patterns (no values)</li>
<li class="check">Domain expert knowledge</li>
</ul>
</div>
</div>
<div class="highlight insight" style="margin-top: 40px;">
<strong>The insight:</strong> You don't need real data. You need a realistic
<em>representation</em> of the business.
</div>
</div>
</div>
<!-- Slide 12: Section - How -->
<div class="slide section-slide">
<div class="section-num">Part 4</div>
<h2>How to Generate</h2>
</div>
<!-- Slide 13: The Process Overview -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">How to Generate</div>
<h2>The 5-Step Process</h2>
</div>
<div class="slide-content top">
<div class="steps">
<div class="step">
<div class="step-num">1</div>
<div class="step-content">
<div class="step-title">Extract & Annotate Schema</div>
<div class="step-desc">Structure + semantics + business context</div>
</div>
</div>
<div class="step">
<div class="step-num">2</div>
<div class="step-content">
<div class="step-title">Extract Usage Patterns</div>
<div class="step-desc">Dashboards, query logs, user interviews</div>
</div>
</div>
<div class="step">
<div class="step-num">3</div>
<div class="step-content">
<div class="step-title">Generate Synthetic Data</div>
<div class="step-desc">Fake but realistic values following patterns</div>
</div>
</div>
<div class="step">
<div class="step-num">4</div>
<div class="step-content">
<div class="step-title">Generate Queries at Scale</div>
<div class="step-desc">AI produces question→SQL pairs</div>
</div>
</div>
<div class="step">
<div class="step-num">5</div>
<div class="step-content">
<div class="step-title">Expand & Validate</div>
<div class="step-desc">Variations, edge cases, human review</div>
</div>
</div>
</div>
</div>
</div>
<!-- Slide 14: Step 1 Detail -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">How to Generate</div>
<h2>Step 1: Extract & Annotate Schema</h2>
</div>
<div class="slide-content top">
<div class="two-col">
<div class="col">
<h3>Input</h3>
<ul>
<li class="bullet">Raw DDL from database</li>
<li class="bullet">Column comments</li>
<li class="bullet">Existing semantic definitions</li>
</ul>
</div>
<div class="col">
<h3>Process</h3>
<ul>
<li class="bullet">LLM-assisted semantic annotation</li>
<li class="bullet">Infer relationships & business rules</li>
<li class="bullet">Domain expert validation</li>
</ul>
</div>
</div>
<h3>Output: Annotated Schema</h3>
<div class="example">
column: amount<br/>
type: decimal<br/>
role: measure<br/>
business_meaning: "Expected deal value in USD"<br/>
typical_range: [1000, 5000000]<br/>
distribution: log_normal<br/>
</div>
</div>
</div>
<!-- Slide 15: Step 2 Detail -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">How to Generate</div>
<h2>Step 2: Extract Usage Patterns</h2>
</div>
<div class="slide-content top">
<div class="two-col">
<div class="col">
<h3>Sources</h3>
<ul>
<li class="bullet">Dashboard SQL (Tableau, Looker, etc.)</li>
<li class="bullet">Query log patterns</li>
<li class="bullet">User interviews</li>
</ul>
</div>
<div class="col">
<h3>Extract</h3>
<ul>
<li class="bullet">Common aggregations & filters</li>
<li class="bullet">Frequent patterns</li>
<li class="bullet">Natural language from titles/labels</li>
</ul>
</div>
</div>
<h3>Output: Query Archetype Catalog</h3>
<div class="example">
pattern: pipeline_by_stage<br/>
archetype: "breakdown"<br/>
natural_language: ["What's our pipeline by stage?",<br/>
"Show me deals in each stage"<br/>,
"Pipeline breakdown"]<br/>
</div>
</div>
</div>
<!-- Slide 16: Steps 3-4 Detail -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">How to Generate</div>
<h2>Steps 3-4: Generate Data & Queries</h2>
</div>
<div class="slide-content top">
<h3>Generate Synthetic Data</h3>
<ul>
<li class="bullet">Realistic fake values following inferred patterns</li>
<li class="bullet">Respect business rules (closed deals have close dates)</li>
<li class="bullet">Inject variation and edge cases deliberately</li>
</ul>
<h3>Generate Queries at Scale</h3>
<ul>
<li class="bullet">LLM generates question→SQL pairs using schema + context</li>
<li class="bullet">Execute every query to verify it runs</li>
<li class="bullet">Human review 10-20% sample</li>
</ul>
<div class="highlight insight" style="margin-top: 24px;">
<strong>Result:</strong> Hundreds of verified pairs in days, instead of weeks.
</div>
</div>
</div>
<!-- Slide 17: Step 5 Detail -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">How to Generate</div>
<h2>Step 5: Expand & Validate</h2>
</div>
<div class="slide-content top">
<h3>Phrasing Variations (3-5 per question)</h3>
<div class="example" style="font-size: 16px;">
Formal: "What is the total revenue by region?"<br/>
Casual: "revenue by region"<br/>
Command: "Show me revenue by region"<br/>
Verbose: "Can you pull up a breakdown of our revenue across regions?"<br/>
</div>
<h3>Edge Cases</h3>
<ul>
<li class="bullet">Ambiguous terms (test clarification handling)</li>
<li class="bullet">Empty results, null handling</li>
<li class="bullet">Date boundaries, fiscal year transitions</li>
</ul>
<h3>Validation</h3>
<ul>
<li class="bullet">Human review catches subtle errors</li>
<li class="bullet">Transfer to production schema for final validation</li>
</ul>
</div>
</div>
<!-- Slide 18: Section - Lessons -->
<div class="slide section-slide">
<div class="section-num">Part 5</div>
<h2>Key Lessons</h2>
</div>
<!-- Slide 19: What Worked -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">Key Lessons</div>
<h2>What Worked</h2>
</div>
<div class="slide-content top">
<ul>
<li class="check">Schema + usage patterns + domain expert = surprisingly good synthetic environment</li>
<li class="check">AI generation with human validation scales 10x faster than manual</li>
<li class="check">Synthetic benchmarks predicted production accuracy within ~10%</li>
<li class="check">Edge cases found in synthetic testing caught real production bugs</li>
</ul>
<div class="highlight insight" style="margin-top: 40px;">
The work is in <strong>understanding the domain</strong>.<br>
The AI generation is the easy part.
</div>
</div>
</div>
<!-- Slide 20: What to Avoid -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">Key Lessons</div>
<h2>What to Avoid</h2>
</div>
<div class="slide-content top">
<h3>Perfectionism on Synthetic Data</h3>
<p>80% realistic is enough. It's scaffolding.</p>
<h3>Skipping Human Validation</h3>
<p>5% error rate in training data compounds into poor system performance.</p>
<h3>Synthetic-Only Evaluation</h3>
<p>Always validate on real data before shipping. Real data has quirks.</p>
<div class="highlight warning" style="margin-top: 32px;">
Synthetic data is for <strong>safe, fast iteration</strong> – not for replacing production validation.
</div>
</div>
</div>
<!-- Slide 21: The Meta-Lesson -->
<div class="slide">
<div class="slide-header">
<div class="slide-number">Key Lessons</div>
<h2>The Meta-Lesson</h2>
</div>
<div class="slide-content">
<div class="big-statement">
Synthetic data enables<br><em>safe, fast iteration</em><br>before you touch production.