1. About me

2. Operating principles

3. Expertise matrix

4. AI & LLM engineering — RAGRAF · AI-Consensus · ctf · LLM Code Projector · MD.Viewer · sr-search-replace · MCP reference

5. Infrastructure engineering — Windows & RDP · email · web/PHP · Linux

6. Tech stack

7. Live metrics

8. Quality bar

9. Documentation & reference

10. Repository navigator (collapsed catalogue below)

11. Collaboration · Connect

👋 About me#

I am Mikhail Deynekin — a senior sysadmin and developer based in Frankfurt am Main,
Germany, in IT continuously since 1991: from DOS workstations and BBS boards to
Kubernetes-era infrastructure and, today, LLM agent tooling. I build the kind of
software I want to operate at 3 a.m.: idempotent, self-documenting, honest about its
failures, and measurable.

• 🔧 35+ years across Windows & Linux infrastructure: automation, hardening, forensics

• 🤖 Current focus: code knowledge graphs for agents (RAGRAF), MCP contracts, multi-model consensus (AI-Consensus)

• 📧 Email infrastructure specialist: hMailServer, Exim, Roundcube, SpamAssassin, SPF/DKIM/DMARC

• 🖥️ RDP security toolkit author and the Windows 11 25H2 enablement script used in production fleets

• 🧾 Docs-as-product: every serious repo ships architecture notes, agent regimens, options references

• 🌐 deynekin.com · 📬 Mikhail@Deynekin.com · 🕐 replies within 24 h

🧭 Operating principles#

🎯 Expertise matrix#

🤖 AI & LLM engineering#

Not a pile of experiments — a connected toolchain. Each tool closes one gap of the
agent-assisted development loop, and every gap I hit in my own work became a repo:

flowchart LR
  ctf["ctf<br/>sources → one<br/>token-budgeted doc"] --> sess(("LLM<br/>session"))
  ragraf["RAGRAF<br/>code knowledge graph<br/>23 MCP tools"] --> sess
  sess --> cons["AI-Consensus<br/>multi-model council:<br/>roles → consensus"]
  cons --> proj["LLM Code Projector<br/>markdown → real<br/>file structures"]
  proj --> view["MD.Viewer<br/>docs, published<br/>beautifully"]
  proj --> sr["sr-search-replace<br/>surgical edits<br/>with rollback"]

🧠 RAGRAF — a graph-and-vector knowledge base of code for LLM agents#

Answers “where is it, who calls it, what breaks if I change it, how is it done
elsewhere” with verifiable file:line facts — instead of blind rg and
full-file reads. One SQLite per corpus holds the graph, the BM25 lexicon, the
vectors and the drift ledger.

Indexing pipeline & stores

flowchart TB
  scan["scan — walk + checksums"] --> ast["ast — tree-sitter: symbols, calls, SQL"]
  ast --> load["load"] --> resolve["resolve — names"]
  resolve --> schema["schema — tables from migrations"]
  schema --> cpg["cpg — Joern interprocedural calls (optional)"]
  cpg --> lexvec["lexvec — FTS5 + vectors + raw chunk text"]
  lexvec --> g[("graph.sqlite<br/>nodes · edges · db schema")]
  lexvec --> l[("lexvec.sqlite<br/>BM25 + vectors + chunks")]
  lexvec --> c[("control.sqlite<br/>revisions · drift ledger · telemetry")]
  g --> mcp{{"23 MCP tools · 22 CLI operations"}}
  l --> mcp
  c --> mcp

Measured, not promised — real PHP portal, 14 navigation tasks, artifacts in bench/results/:

Three design decisions, each born from a measured defect

1. Immutable index + drift ledger instead of “update on the fly”. Inside a revision the index never changes; code edits accumulate in the ledger and affected facts turn stale. Mixing timestamps inside one answer is exactly how wrong conclusions are born.

2. No free-form graph query language for the agent. Fixed operations only: in LLMxCPG (USENIX Security ’25) base models produced 132/1278 and 19/1278 valid CPGQL queries versus 1278/1278 for a fine-tuned one — a query language turns base models into error generators.

3. Every fact carries proof: path:line, provenance (stated / derived), index_rev, confidence. An edge without evidence looks credible only because it came “from the database”.

There is, and will be, no free-form graph query tool (FT-15): a MATCH … RETURN
attempt is rejected with code -32602.

Quick start

ragraf init <project-dir>                       # creates config/myproject.yaml
python tools/ragraf.py --project config/myproject.yaml index

{
  "mcpServers": {
    "ragraf": {
      "command": "python",
      "args": ["<repo>/tools/ragraf-mcp.py", "--project", "<repo>/config/myproject.yaml"]
    }
  }
}

🏛️ AI-Consensus — an OpenCorum-style council of LLMs#

A PHP 8.4 web application where several models deliberate in defined roles, a
consensus engine scores their agreement, and every answer ships a clickable DAG of the
whole session. Full audit trail: every model response, every step, every score — in MySQL.

flowchart TB
  q["User query"] --> l0["L0 · Skill Router — pick the matching SKILL,<br/>or auto-create one from the query"]
  l0 --> l1["L1 · Query distribution — parallel calls<br/>to multiple roles × models (BotHub, OpenAI-compatible)"]
  l1 --> l2["L2 · Response normalization —<br/>JSON-schema enforcement per role"]
  l2 --> l3["L3 · Consensus analysis —<br/>agreement scores, conflict detection"]
  l3 --> ans["Answer + interactive “Details” DAG"]
  mem[("Vector memory:<br/>short & long term")] -.-> l1
  aud[("MySQL 8 audit trail:<br/>PDO, prepared, transactions")] -.-> l3

• SKILLS system — declarative JSON workflows defining agent roles, prompts and consensus rules; the router auto-selects or auto-creates the most relevant SKILL per query.

• Strict PHP 8.4 — strict types, readonly properties, enums throughout; six-part design doc from architecture to the full CouncilEngine module.

ctf — collect-to-file: context assembly with a token budget#

One self-contained POSIX script (bash 4.2+, shellcheck-clean) that folds a source tree
into a single Markdown / JSON / JSONL / text document with relative paths — the
document an LLM reasons over best, or a reviewer reads in one pass.

ctf --budget 60000 src/ > context.md        # LLM context, capped at a token budget
ctf --format jsonl --hashes . > audit.jsonl  # auditable snapshot, per-file hashes

It does not shell out to eval, does not execute anything it reads, never writes
outside the file you asked for.

🏗️ LLM Code Projector — markdown answers → real file structures#

The reverse direction of the loop: feed it a ChatGPT/Claude/DeepSeek/Gemini answer with
project trees and code fences — get a runnable directory tree in one command.
Recognizes ASCII and Unicode trees, any nesting depth.

./MD_to_Files.sh chatgpt_response.md ./my-new-project

📚 MD.Viewer — zero-build documentation publishing#

The loop’s last mile: LLM- and human-written markdown rendered as polished docs.
Auto TOC with heading numbering, Mermaid with auto-repair, glossary tooltips, dark mode,
code-copy controls, searchable file browser, built-in self-updater — no build step,
no database, no framework: copy md.php next to your .md files and open it.

🔧 sr-search-replace — the “apply the fix” step#

Universal search & replace with session rollback, binary detection and backup
management: the safe way to let an agent (or yourself) touch forty files at once.

🏭 Infrastructure engineering#

The other half of my life: systems that must stay up, stay secure, and stay boring.

🪟 Windows & RDP security#

.\Upgrade-Win11-To-25H2.ps1 -RetryCount 5    # enterprise guardrails baked in

📬 Email infrastructure — the full delivery chain#

flowchart LR
  net((Internet)) --> mx["hMailServer / Exim<br/>MX + TLS"]
  mx --> sa["SpamAssassin (spamd)<br/>daily rule updates"]
  sa --> mb["Mailboxes<br/>idempotent SPAM provisioning"]
  mb --> rc["Roundcube<br/>IIS + PHP FastCGI"]
  rc --> pg[("PostgreSQL<br/>prefs & cache")]
  le["cert-unifier · Let’s Encrypt"] -. TLS .-> mx
  le -. TLS .-> rc

• hMailServer at the core: domains, accounts, aliases, lists, CSV/JSON export, antivirus integration (Defender, Kaspersky, ClamAV)

• SpamAssassin as a Windows service under PowerShell 7 with Task-Scheduler rule updates

• Deliverability stack: SPF · DKIM · DMARC configured and monitored, not “set and forgotten”

• Roundcube ops: deployment on IIS + FastCGI, PostgreSQL backend, cache truncation tooling

🌐 Web, PHP & databases#

• HTML.PHP — single-file PHP 8.3+ template engine: Apache SSI directives, html4.pl-compatible API, XSS-safe by default

• IIS-PHP-FastCGI-AutoDeploy — one-shot IIS + PHP FastCGI installer (Roundcube-ready)

• pg-encoding-migrator — safe PostgreSQL encoding migration (e.g. WIN1251 → UTF-8)

• woocommerce-1c-data-cleaner · Bash_WP-CLI_Update · Clean_BOM_Senior · BAS-IP

🐧 Linux & network ops#

• sata-spindown-guard — cron-based HDD spindown guard (hdparm), safe for unmounted drives

• cert-unifier — unified PEM (key + chain) for Let’s Encrypt deployments

• RouterOS filtering & monitoring practice; bash tooling with cron discipline and dry-run habits

🧰 Tech stack#

Languages & scripting

AI & LLM

Infrastructure & OS

Databases

Email & web

📊 Live metrics#

Stats & language cards are local SVG snapshots rebuilt from the live GitHub API
by build_svg_assets.py (delivery package) — immune to third-party rate limits and outages;
streak stays live via streak-stats.

🏅 Quality bar#

📚 Documentation & reference#

Every serious repo ships its own field guide. The most useful entry points:

AI & LLM
RAGRAF · AI-Consensus · ctf-collect-to-file · LLM Code Projector · MD.Viewer · sr-search-replace

Windows & RDP security
Windows-11-25H2-Update-Script · RDP-Diagnostic-Tool · 1st-Remote-Session-Manager-Pro · 1st-RDPMon · RDPAudit · Get-Windows-Security-Events-By-IP · Windows-LogOff · windows-show-all-tray-icons · Windows-Cleaner-and-Optimizer · Clear-TempFolders · Print-Spooler-Queue-Cleanup · Import-CaCertificates

Email infrastructure
hMailServer-Spam-Provisioner · SpamAssassin-Windows-PS7 · PowerShell_Export-hMailServerData · hMailServer-External-Antivirus · hmailserver-kaspersky-cleanup · hmailserver-clamav-cleanup · Clean-RoundCube-Cache-PostgreSQL-Windows · Contact-Form-7-Spam-Checker

Web, PHP & databases
HTML.PHP · IIS-PHP-FastCGI-AutoDeploy · Bash_WP-CLI_Update · woocommerce-1c-data-cleaner · BAS-IP · Clean_BOM_Senior · pg-encoding-migrator

Linux & certificates
sata-spindown-guard · cert-unifier

• hMailServer — domains, accounts, aliases, distribution lists, CSV/JSON export, antivirus integration (Defender, Kaspersky, ClamAV)

• SpamAssassin on Windows — spamd as a system service under PowerShell 7 with daily rule updates via Task Scheduler

• Idempotent provisioning — [SPAM] IMAP folders + filter rules for every account, safe to re-run

• Virus cleanup — Kaspersky and ClamAV scan-and-quarantine tooling for the hMailServer store

• Deliverability stack — SPF · DKIM · DMARC configuration and hygiene

• Roundcube — deployment on IIS + PHP FastCGI, PostgreSQL backend, cache truncation tooling

• Exim — routing, filtering, delivery tuning on Linux

• RDP-Diagnostic-Tool — full readiness check & auto-remediation

• 1st-RDPMon — real-time server-side session monitoring & alerting

• 1st-Remote-Session-Manager-Pro — shadow, control, disconnect, message, logoff

• RDPAudit — security monitoring & session auditing (C#)

• Get-Windows-Security-Events-By-IP — brute-force & auth-failure forensics

• Windows-LogOff — forcible logoff with residual-process cleanup

• Windows-11-25H2-Update-Script — enablement-package upgrade with BITS fallback & signature validation

• Windows-Cleaner-and-Optimizer — safe browser/system cache cleanup

• Clear-TempFolders — TEMP hygiene with DryRun mode

• Print-Spooler-Queue-Cleanup — forcible spooler queue clear with admin checks

• Import-CaCertificates — CA root/intermediate import into Windows stores

• windows-show-all-tray-icons — notification-area registry tweak

🤝 Collaboration#

Where I add value fastest:

• 📧 Email deliverability rescues — SPF/DKIM/DMARC alignment, spam-filter tuning, hMailServer/Exim migrations

• 🖥️ RDP hardening & session operations for Windows fleets, from diagnostics to forensics

• 🪟 Windows upgrade/update automation with enterprise guardrails (signatures, retries, logging)

• 🤖 LLM agent tooling — code knowledge bases, MCP contracts, context pipelines, consensus patterns

• 🧾 Evidence-based codebase audits — RAGRAF-driven, every finding with file:line

Open-source collaboration is welcome on any repo: issues with reproduction steps get
answers, pull requests with tests get merged.

🧭 Currently#

• 🔭 Building: RAGRAF — making code knowledge bases a first-class citizen for LLM agents

• 🌱 Exploring: agent-safe MCP contracts, consensus scoring for multi-model councils

• 👯 Open to: collaborations on AI tooling · consulting on email & RDP infrastructure

• 📫 Reach me: Mikhail@Deynekin.com — I reply within 24 hours

🌐 Connect#

Infrastructure that survives audits. AI tooling that survives reality.