-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
669 lines (615 loc) · 33.5 KB
/
Copy pathindex.html
File metadata and controls
669 lines (615 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
<!DOCTYPE html>
<html lang="en" prefix="og: https://ogp.me/ns#">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>DrobyshevDev — frameworks and tooling for ML, LLM agents and operational decisions</title>
<meta name="description" content="DrobyshevDev builds open-source frameworks and tooling for machine learning, LLM agents and operational decision systems: praxis, mlango, glia, decisionrl and stadion. Typed, tested in CI on Linux, macOS and Windows.">
<link rel="canonical" href="https://drobyshevdev.github.io/">
<link rel="alternate" hreflang="en" href="https://drobyshevdev.github.io/">
<link rel="alternate" hreflang="ru" href="https://drobyshevdev.github.io/ru/">
<link rel="alternate" hreflang="x-default" href="https://drobyshevdev.github.io/">
<meta name="theme-color" content="#fdfcfa" media="(prefers-color-scheme: light)">
<meta name="theme-color" content="#131316" media="(prefers-color-scheme: dark)">
<meta property="og:type" content="website">
<meta property="og:site_name" content="DrobyshevDev">
<meta property="og:locale" content="en_US">
<meta property="og:locale:alternate" content="ru_RU">
<meta property="og:title" content="DrobyshevDev — frameworks and tooling for ML, LLM agents and operational decisions">
<meta property="og:description" content="Open-source frameworks and tooling for machine learning, LLM agents and operational decision systems. Typed, tested on Linux, macOS and Windows.">
<meta property="og:url" content="https://drobyshevdev.github.io/">
<meta property="og:image" content="https://drobyshevdev.github.io/assets/og.png">
<meta property="og:image:width" content="1200">
<meta property="og:image:height" content="630">
<meta property="og:image:alt" content="DrobyshevDev — frameworks and tooling for machine learning and LLM agents. praxis, mlango, glia, decisionrl, stadion.">
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:image" content="https://drobyshevdev.github.io/assets/og.png">
<link rel="icon" href="/assets/favicon.png" type="image/png">
<link rel="apple-touch-icon" href="/assets/apple-touch-icon.png">
<link rel="stylesheet" href="/assets/site.css">
<script>
// Applied before first paint so the page never flashes the wrong theme.
try {
var t = localStorage.getItem("theme");
if (t === "dark" || t === "light") document.documentElement.setAttribute("data-theme", t);
} catch (e) {}
</script>
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "Organization",
"@id": "https://drobyshevdev.github.io/#org",
"name": "DrobyshevDev",
"url": "https://drobyshevdev.github.io/",
"logo": "https://drobyshevdev.github.io/assets/mark.png",
"image": "https://drobyshevdev.github.io/assets/og.png",
"description": "Open-source frameworks and tooling for machine learning, LLM agents and operational decision systems.",
"sameAs": ["https://github.com/DrobyshevDev"],
"contactPoint": {
"@type": "ContactPoint",
"contactType": "business enquiries",
"email": "denidrob138@gmail.com",
"availableLanguage": ["en", "ru"]
}
},
{
"@type": "SoftwareSourceCode",
"name": "praxis",
"description": "A legal assistant for Russian law that answers with the provisions it rests on, every citation verified by an NLI model.",
"codeRepository": "https://github.com/DrobyshevDev/praxis",
"programmingLanguage": "Python",
"license": "https://www.apache.org/licenses/LICENSE-2.0",
"author": { "@id": "https://drobyshevdev.github.io/#org" }
},
{
"@type": "SoftwareSourceCode",
"name": "mlango",
"description": "A batteries-included framework for machine learning, analytics and LLM agents, built on Django's philosophy.",
"codeRepository": "https://github.com/DrobyshevDev/mlango",
"programmingLanguage": "Python",
"license": "https://opensource.org/licenses/MIT",
"author": { "@id": "https://drobyshevdev.github.io/#org" }
},
{
"@type": "SoftwareSourceCode",
"name": "glia",
"description": "A glass-box, minimal library for building LLM agents: modern techniques as opt-in primitives, no hidden control flow.",
"codeRepository": "https://github.com/DrobyshevDev/glia",
"programmingLanguage": "Python",
"license": "https://opensource.org/licenses/MIT",
"author": { "@id": "https://drobyshevdev.github.io/#org" }
},
{
"@type": "SoftwareSourceCode",
"name": "decisionrl",
"description": "Reinforcement learning for operational decisions, each applied environment shipped with its classical operations-research baseline.",
"codeRepository": "https://github.com/DrobyshevDev/decisionrl",
"programmingLanguage": "Python",
"license": "https://opensource.org/licenses/MIT",
"author": { "@id": "https://drobyshevdev.github.io/#org" }
},
{
"@type": "SoftwareSourceCode",
"name": "stadion",
"description": "A proving ground for operational decisions: agents scored against the tuned classical method and the exact optimum, with a confidence interval.",
"codeRepository": "https://github.com/DrobyshevDev/stadion",
"programmingLanguage": "Python",
"license": "https://opensource.org/licenses/MIT",
"author": { "@id": "https://drobyshevdev.github.io/#org" }
},
{
"@type": "Course",
"name": "lemma",
"description": "A free course covering machine learning, neural networks, reinforcement learning and recommender systems, from zero to reading and reproducing research.",
"url": "https://drobyshevdev.github.io/lemma/",
"inLanguage": "ru",
"isAccessibleForFree": true,
"provider": { "@id": "https://drobyshevdev.github.io/#org" }
}
]
}
</script>
</head>
<body>
<a class="skip-link" href="#main">Skip to content</a>
<header class="site-header">
<div class="wrap">
<a class="brand" href="/">
<img class="mark" src="/assets/mark.png" alt="" width="24" height="24">
DrobyshevDev
</a>
<nav class="nav" id="primary-nav" aria-label="Primary">
<a href="#projects">Projects</a>
<a href="#stack">Stack</a>
<a href="#evidence">Evidence</a>
<a href="#notes">Notes</a>
<a href="#method">Method</a>
<a href="https://github.com/DrobyshevDev" rel="noopener">GitHub</a>
</nav>
<div class="nav-tools">
<a class="lang-switch" href="/ru/" hreflang="ru" lang="ru">RU</a>
<button class="icon-btn theme-toggle" type="button"
data-label-dark="Switch to the light theme"
data-label-light="Switch to the dark theme"
aria-label="Switch colour theme">
<svg class="sun" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" aria-hidden="true"><circle cx="12" cy="12" r="4"/><path d="M12 2v2M12 20v2M4.9 4.9l1.4 1.4M17.7 17.7l1.4 1.4M2 12h2M20 12h2M4.9 19.1l1.4-1.4M17.7 6.3l1.4-1.4"/></svg>
<svg class="moon" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M21 12.8A9 9 0 1 1 11.2 3a7 7 0 0 0 9.8 9.8z"/></svg>
</button>
<button class="icon-btn nav-toggle" type="button" aria-expanded="false" aria-controls="primary-nav" aria-label="Open menu">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" aria-hidden="true"><path d="M4 7h16M4 12h16M4 17h16"/></svg>
</button>
</div>
</div>
</header>
<main id="main">
<!-- ==================== Hero ==================== -->
<section class="hero">
<div class="wrap">
<div class="hero-grid">
<div class="hero-copy">
<p class="eyebrow">Open source · Python · MIT and Apache-2.0</p>
<h1>Frameworks and tooling for machine learning, LLM agents and operational decisions.</h1>
<p class="lead">
The layer between a research result and something a team can run: the project structure,
the tracking, the interfaces, and the tests that keep a number honest six months after it
was measured.
</p>
<div class="actions">
<a class="btn btn-bright" href="#projects">
See the projects
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg>
</a>
<a class="btn btn-secondary" href="https://github.com/DrobyshevDev" rel="noopener">
<svg viewBox="0 0 24 24" fill="currentColor" aria-hidden="true"><path d="M12 .3a12 12 0 0 0-3.8 23.4c.6.1.8-.3.8-.6v-2c-3.3.7-4-1.6-4-1.6-.6-1.4-1.4-1.8-1.4-1.8-1.1-.7.1-.7.1-.7 1.2.1 1.8 1.2 1.8 1.2 1.1 1.8 2.8 1.3 3.5 1 .1-.8.4-1.3.8-1.6-2.7-.3-5.5-1.3-5.5-5.9 0-1.3.5-2.4 1.2-3.2-.1-.3-.5-1.5.1-3.2 0 0 1-.3 3.3 1.2a11.5 11.5 0 0 1 6 0C17.3 4.8 18.3 5.1 18.3 5.1c.6 1.7.2 2.9.1 3.2.8.8 1.2 1.9 1.2 3.2 0 4.6-2.8 5.6-5.5 5.9.4.4.8 1.1.8 2.2v3.3c0 .3.2.7.8.6A12 12 0 0 0 12 .3z"/></svg>
GitHub
</a>
</div>
<p class="hero-since"><span class="dot"></span> Five libraries, a free course and open reading notes. Since 2026.</p>
</div>
<div class="hero-mark">
<button class="mark-trigger" type="button"
aria-expanded="false" aria-controls="mark-menu" aria-label="Open the project menu">
<img src="/assets/mark.png" alt="" width="372" height="372" decoding="async">
</button>
<div class="mark-menu" id="mark-menu" hidden>
<p class="mark-menu-title">Go straight to</p>
<ul role="list">
<li><a href="https://drobyshevdev.github.io/praxis/"><span class="mark-menu-name">praxis</span><span class="mark-menu-note">documentation</span></a></li>
<li><a href="https://drobyshevdev.github.io/mlango/"><span class="mark-menu-name">mlango</span><span class="mark-menu-note">documentation</span></a></li>
<li><a href="https://drobyshevdev.github.io/glia/"><span class="mark-menu-name">glia</span><span class="mark-menu-note">documentation</span></a></li>
<li><a href="https://drobyshevdev.github.io/decisionrl/"><span class="mark-menu-name">decisionrl</span><span class="mark-menu-note">documentation</span></a></li>
<li><a href="https://drobyshevdev.github.io/stadion/"><span class="mark-menu-name">stadion</span><span class="mark-menu-note">documentation</span></a></li>
<li><a href="https://drobyshevdev.github.io/lemma/"><span class="mark-menu-name">lemma</span><span class="mark-menu-note">the course</span></a></li>
<li><a href="https://github.com/DrobyshevDev"><span class="mark-menu-name">GitHub</span><span class="mark-menu-note">every repository</span></a></li>
</ul>
</div>
</div>
</div>
</div>
</section>
<!-- ==================== Projects ==================== -->
<section class="band" id="projects">
<div class="wrap">
<div class="section-head">
<p class="eyebrow">Projects</p>
<h2>What we build</h2>
<p class="lead">
Each one installs today, carries documentation and is released under a permissive
licence. The numbers beside each project come from a script in the repository that
produced them.
</p>
</div>
<div class="cards">
<article class="card">
<div class="card-top">
<h3><a href="https://github.com/DrobyshevDev/praxis" rel="noopener">praxis</a></h3>
<span class="tag">Apache-2.0</span>
</div>
<p>
A legal assistant whose citations are checked, not asserted. Hybrid retrieval with a
cross-encoder reranker, and an NLI model that verifies each cited provision supports
the claim. The default answer is the text of the law and cannot hallucinate.
</p>
<div class="command"><span class="p">$ </span>docker compose up app</div>
<div class="card-foot">
<a class="more" href="https://drobyshevdev.github.io/praxis/" rel="noopener">Documentation
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
<a class="more" href="https://github.com/DrobyshevDev/praxis" rel="noopener">Repository
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
</div>
</article>
<article class="card">
<div class="card-top">
<h3><a href="https://github.com/DrobyshevDev/mlango" rel="noopener">mlango</a></h3>
<span class="tag">MIT</span>
</div>
<p>
Django's philosophy applied to machine learning. You declare datasets, models, agents
and evaluations; one class body becomes an admin page, a documented endpoint, a
migration and a CLI target at once.
</p>
<div class="command"><span class="p">$ </span>pip install "mlango[sklearn]"</div>
<div class="card-foot">
<a class="more" href="https://drobyshevdev.github.io/mlango/" rel="noopener">Documentation
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
<a class="more" href="https://pypi.org/project/mlango/" rel="noopener">PyPI
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
</div>
</article>
<article class="card">
<div class="card-top">
<h3><a href="https://github.com/DrobyshevDev/glia" rel="noopener">glia</a></h3>
<span class="tag">MIT</span>
</div>
<p>
An agent loop you can read in an afternoon. Every model call, tool call and state
transition is a plain object you can log, snapshot and replay. The core imports
nothing outside the standard library.
</p>
<div class="command"><span class="p">$ </span>pip install glia-agents</div>
<div class="card-foot">
<a class="more" href="https://drobyshevdev.github.io/glia/" rel="noopener">Documentation
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
<a class="more" href="https://pypi.org/project/glia-agents/" rel="noopener">PyPI
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
</div>
</article>
<article class="card">
<div class="card-top">
<h3><a href="https://github.com/DrobyshevDev/decisionrl" rel="noopener">decisionrl</a></h3>
<span class="tag">MIT</span>
</div>
<p>
Reinforcement learning for decisions that have a cost function: pricing, inventory,
energy, queues, supply chains. Every applied environment ships the classical
operations-research baseline beside it.
</p>
<div class="command"><span class="p">$ </span>pip install decisionrl</div>
<div class="card-foot">
<a class="more" href="https://drobyshevdev.github.io/decisionrl/" rel="noopener">Documentation
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
<a class="more" href="https://pypi.org/project/decisionrl/" rel="noopener">PyPI
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
</div>
</article>
<article class="card">
<div class="card-top">
<h3><a href="https://github.com/DrobyshevDev/stadion" rel="noopener">stadion</a></h3>
<span class="tag">MIT</span>
</div>
<p>
A proving ground where an agent is scored against the exact optimum. Six operational
decisions, each with the tuned classical method and the computed optimum beside it,
and a bootstrap interval on the difference.
</p>
<div class="command"><span class="p">$ </span>pip install stadion-rl</div>
<div class="card-foot">
<a class="more" href="https://drobyshevdev.github.io/stadion/" rel="noopener">Documentation
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
<a class="more" href="#evidence">The measurements
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
<a class="more" href="https://github.com/DrobyshevDev/stadion" rel="noopener">Repository
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
</div>
</article>
</div>
</div>
</section>
<!-- ==================== Stack ==================== -->
<section class="band band-2" id="stack">
<div class="wrap">
<div class="section-head">
<p class="eyebrow">Stack</p>
<h2>What these are built on, and why</h2>
<p class="lead">
A short list on purpose. Every dependency is something a reader has to install, learn
and keep updated, so each one here earns its place in more than one project.
</p>
</div>
<div class="cards">
<article class="card">
<div class="card-top"><h3>Python</h3><span class="tag">3.9 – 3.13</span></div>
<p>
Typed throughout, checked with mypy and linted with ruff in CI. The version floors are
lower bounds on what a user may already have, not the version CI runs — raising them
would lock people out without testing anything new.
</p>
</article>
<article class="card">
<div class="card-top"><h3>PyTorch</h3><span class="tag">decisionrl</span></div>
<p>
Every algorithm in decisionrl. CPU is the default target: the applied environments are
small enough that a reader without a GPU can still reproduce the numbers in the
documentation, which is the point of publishing them.
</p>
</article>
<article class="card">
<div class="card-top"><h3>FastAPI · NumPy</h3><span class="tag">praxis</span></div>
<p>
One API core serves every client. Dense retrieval is a matrix-vector product over
vectors held in memory and cached to disk, with a pure standard-library fallback so
the offline image and CI run without NumPy at all. A Postgres index with pgvector is
designed and not yet wired — the schema is in the repository, the retriever is not.
</p>
</article>
<article class="card">
<div class="card-top"><h3>Docker</h3><span class="tag">praxis</span></div>
<p>
The image installs the API extra, sets offline mode and carries the corpus, so
<code>docker compose up app</code> runs with no key, no GPU and no outbound request.
A reader who cannot start it has no way to check anything.
</p>
</article>
<article class="card">
<div class="card-top"><h3>GitHub Actions</h3><span class="tag">every repository</span></div>
<p>
Linux, macOS and Windows on every push, with one aggregate check to require in branch
protection. Releases publish through Trusted Publishing, and the desktop binaries carry
a signed provenance attestation.
</p>
</article>
<article class="card">
<div class="card-top"><h3>MkDocs</h3><span class="tag">documentation</span></div>
<p>
Built with <code>--strict</code>, which fails on a broken internal link or a page
missing from the navigation. No webfonts and no analytics on any of the sites,
including this one.
</p>
</article>
</div>
</div>
</section>
<!-- ==================== Evidence ==================== -->
<section class="band band-bright" id="evidence">
<div class="wrap">
<div class="section-head">
<p class="eyebrow">Evidence</p>
<h2>Numbers with the conditions they were measured under</h2>
<p class="lead">
There are no client logos here and no case studies. What an open-source organisation can
show instead is its measurements, and the commands that reproduce them.
</p>
</div>
<div class="measure">
<table>
<caption>
Headroom between the tuned classical method and the computed optimum in stadion, with a
95% bootstrap interval on the paired difference. Measured over 40 instances × 20
episodes; reproduce with <code>stadion run <task> --agent classical --instances 40 --episodes 20</code>.
</caption>
<thead>
<tr><th>Task</th><th>Headroom</th><th>95% interval</th></tr>
</thead>
<tbody>
<tr><td><code>inventory</code></td><td>+0.4%</td><td>[+0.561, +1.031]</td></tr>
<tr><td><code>joint-pricing</code></td><td>+4.3%</td><td>[+3.794, +5.656]</td></tr>
<tr><td><code>pricing</code></td><td>+6.2%</td><td>[+1.265, +1.761]</td></tr>
<tr><td><code>queueing</code></td><td>+16.9%</td><td>[+3.418, +3.988]</td></tr>
<tr><td><code>supply-chain</code></td><td>+17.3%</td><td>[+5.505, +7.493]</td></tr>
<tr><td><code>energy</code></td><td>+26.6%</td><td>[+4.221, +4.690]</td></tr>
</tbody>
</table>
</div>
<p class="lead" style="margin-top:28px">
The spread is the point. Where the textbook answer is already near-optimal there is almost
nothing to win; where it cannot plan ahead, a quarter of the value is still on the table. A
benchmark whose tasks all leave generous headroom has selected for problems the standard
method happens to be bad at.
</p>
<dl class="figures">
<div class="figure">
<dt>Algorithms</dt>
<dd>31<span class="sub">In decisionrl, across 22 environments — nine of them applied rather than synthetic.</span></dd>
</div>
<div class="figure">
<dt>Required dependencies</dt>
<dd>0<span class="sub">The glia core imports nothing outside the standard library. Providers and extras are opt-in.</span></dd>
</div>
<div class="figure">
<dt>Line coverage</dt>
<dd>86%<span class="sub">decisionrl, measured by pytest-cov in CI. The build fails below the threshold rather than reporting it.</span></dd>
</div>
<div class="figure">
<dt>Retrieval recall@5</dt>
<dd>0.92<span class="sub">praxis, on the golden set over the full corpus, tracked run over run. MRR 0.94.</span></dd>
</div>
</dl>
</div>
</section>
<!-- ==================== Notes ==================== -->
<section class="band" id="notes">
<div class="wrap">
<div class="section-head">
<p class="eyebrow">Notes and teaching</p>
<h2>Written to be checked</h2>
<p class="lead">
A paper says what it found. A note says what it showed, which is a smaller thing, and
marks the distance between them.
</p>
</div>
<div class="feed">
<article class="card">
<p class="meta"><time datetime="2026-08-10">10 August 2026</time><span class="sep">·</span>research<span class="sep">·</span>draft</p>
<h3><a href="https://drobyshevdev.github.io/research/notes/dopamine-reward-prediction-error/" rel="noopener">Dopamine neurons and the reward prediction error</a></h3>
<p>
Midbrain dopamine neurons fire in a pattern that looks like the error term of
temporal-difference learning. Close enough that an algorithm from 1988 predicted what
an electrode recorded in 1997 — which is why the result gets overstated.
</p>
<div class="chips"><span class="chip">neuroscience</span><span class="chip">reinforcement-learning</span><span class="chip">dopamine</span></div>
</article>
<article class="card">
<p class="meta"><span>Open collection</span><span class="sep">·</span>CC BY 4.0</p>
<h3><a href="https://drobyshevdev.github.io/research/" rel="noopener">research — open reading notes</a></h3>
<p>
Machine learning, reinforcement learning, neuroscience and the methods used to check a
claim. Every note declares the sources it rests on, and CI rejects one that does not.
</p>
<div class="card-foot" style="margin-top:16px">
<a class="more" href="https://drobyshevdev.github.io/research/" rel="noopener">Read the notes
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
</div>
</article>
<article class="card">
<p class="meta"><span>Free course</span><span class="sep">·</span>27 modules<span class="sep">·</span><span lang="ru">Русский</span></p>
<h3><a href="https://drobyshevdev.github.io/lemma/" rel="noopener">lemma — the road into ML, DL and RL</a></h3>
<p>
From the arithmetic of a mean to reproducing a recent paper. No sign-up and no first
module free. Module one is about baselines and confidence intervals, before any machine
learning at all, because the central skill is checking a claim rather than launching a
training run.
</p>
<div class="card-foot" style="margin-top:16px">
<a class="more" href="https://drobyshevdev.github.io/lemma/" rel="noopener">Open the course
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M5 12h14M13 6l6 6-6 6"/></svg></a>
</div>
</article>
</div>
</div>
</section>
<!-- ==================== Method ==================== -->
<section class="band band-dark" id="method">
<div class="wrap">
<div class="section-head">
<p class="eyebrow">Method</p>
<h2>Five rules every repository is held to</h2>
<p class="lead">
Review criteria, not aspirations. A pull request that breaks one of them gets comments.
</p>
</div>
<div class="rules">
<div class="rule">
<span class="num">01</span>
<h3>Errors teach</h3>
<p>
A message is read at the worst possible moment by someone who does not have the source
open. It should say what went wrong and what to do next, and list the alternatives when
there are any.
</p>
</div>
<div class="rule">
<span class="num">02</span>
<h3>Tests are named after the guarantee they protect</h3>
<p>
<code>test_assignment_is_stable_when_rows_are_added</code>, not <code>test_split</code>.
A test whose name does not survive being read aloud is not documenting anything.
</p>
</div>
<div class="rule">
<span class="num">03</span>
<h3>Verification beats assertion</h3>
<p>
A green pipeline on one machine is not evidence. Every published claim is measured on
the run it describes, and the numbers in a README are pinned by tests, because prose
rots quietly and a failing test does not.
</p>
</div>
<div class="rule">
<span class="num">04</span>
<h3>Comments explain why</h3>
<p>
The code already says what it does. The comment is for the constraint a reader cannot
see.
</p>
</div>
<div class="rule">
<span class="num">05</span>
<h3>No hidden control flow</h3>
<p>
If a framework does something on your behalf, it should be possible to find the line
where it happens.
</p>
</div>
</div>
</div>
</section>
<!-- ==================== Licensing ==================== -->
<section class="band" id="licensing">
<div class="wrap">
<div class="section-head">
<p class="eyebrow">Working together</p>
<h2>Terms, in one place</h2>
</div>
<dl class="rows">
<div>
<dt>Licences</dt>
<dd>mlango, glia, decisionrl and stadion are MIT. praxis is Apache-2.0. Course and research prose is CC BY 4.0, its code MIT. Commercial use is permitted under all of them.</dd>
</div>
<div>
<dt>Commercial enquiries</dt>
<dd>Integration, a feature your team needs upstream, or work on something adjacent — <a href="mailto:denidrob138@gmail.com?subject=DrobyshevDev%20—%20enquiry">denidrob138@gmail.com</a>. Say what you are building and what is in the way; that is enough for a first reply.</dd>
</div>
<div>
<dt>Contributing</dt>
<dd>Open an issue describing the change, then a pull request against the default branch. Every repository carries a <code>CONTRIBUTING.md</code> — its own or the organisation-wide default — with the checks CI will run, so there is no unwritten rule to discover afterwards.</dd>
</div>
<div>
<dt>Reporting a vulnerability</dt>
<dd>Use GitHub's private vulnerability reporting on the affected repository. Never a public issue.</dd>
</div>
<div>
<dt>Code of conduct</dt>
<dd>Contributor Covenant, organisation-wide. Reports go privately to the address above.</dd>
</div>
</dl>
</div>
</section>
</main>
<footer class="site-footer">
<div class="wrap">
<div class="footer-grid">
<div class="footer-brand">
<a class="brand" href="/">
<img class="mark" src="/assets/mark.png" alt="" width="24" height="24">
DrobyshevDev
</a>
<p>
Open-source frameworks and tooling for machine learning, LLM agents and operational
decision systems.
</p>
</div>
<div class="footer-col">
<h4>Projects</h4>
<ul role="list">
<li><a href="https://github.com/DrobyshevDev/praxis" rel="noopener">praxis</a></li>
<li><a href="https://github.com/DrobyshevDev/mlango" rel="noopener">mlango</a></li>
<li><a href="https://github.com/DrobyshevDev/glia" rel="noopener">glia</a></li>
<li><a href="https://github.com/DrobyshevDev/decisionrl" rel="noopener">decisionrl</a></li>
<li><a href="https://github.com/DrobyshevDev/stadion" rel="noopener">stadion</a></li>
</ul>
</div>
<div class="footer-col">
<h4>Documentation</h4>
<ul role="list">
<li><a href="https://drobyshevdev.github.io/praxis/" rel="noopener">praxis</a></li>
<li><a href="https://drobyshevdev.github.io/mlango/" rel="noopener">mlango</a></li>
<li><a href="https://drobyshevdev.github.io/glia/" rel="noopener">glia</a></li>
<li><a href="https://drobyshevdev.github.io/decisionrl/" rel="noopener">decisionrl</a></li>
<li><a href="https://drobyshevdev.github.io/stadion/" rel="noopener">stadion</a></li>
<li><a href="https://drobyshevdev.github.io/lemma/" rel="noopener">lemma, the course</a></li>
<li><a href="https://drobyshevdev.github.io/research/" rel="noopener">research</a></li>
</ul>
</div>
<div class="footer-col">
<h4>Organisation</h4>
<ul role="list">
<li><a href="https://github.com/DrobyshevDev" rel="noopener">GitHub</a></li>
<li><a href="#method">Method</a></li>
<li><a href="#licensing">Licensing</a></li>
<li><a href="mailto:denidrob138@gmail.com?subject=DrobyshevDev%20—%20enquiry">Contact</a></li>
<li><a href="/ru/" hreflang="ru" lang="ru">Русская версия</a></li>
<li><a href="/legal/" hreflang="ru">Legal <span lang="en">(in Russian)</span></a></li>
</ul>
</div>
</div>
<div class="footer-bottom">
<p>© <span data-year>2026</span> DrobyshevDev. Source under MIT and Apache-2.0.</p>
<p>Static site. No trackers, no third-party requests — <a href="https://github.com/DrobyshevDev/DrobyshevDev.github.io/blob/main/scripts/check_site.py" rel="noopener">checked in CI</a>, not asserted.</p>
</div>
</div>
</footer>
<script src="/assets/site.js" defer></script>
</body>
</html>