3–5 business days
Technical assessment
Review of the architecture, critical path and constraints.
- risk and bottleneck map
- solution options and trade-offs
- validation and next-stage plan
- written findings
Independent engineering practice · kernels · telecom · storage
Systems software and technical advisory for complex infrastructure products
KernelSoft is Maxim Lyadvinsky’s independent practice. I personally build critical systems components and conduct independent reviews of R&D, architecture and technical intellectual property.
ENGINEERINGDrivers, data planes, storage and stabilizationDevelopment and optimization of critical components.
ADVISORYR&D due diligence, startup rating and technical IPIndependent judgment before an investment, acquisition or major product transition.
One accountable expertAdditional specialists only with customer approvalConfidentiality agreement; work in the customer repository and CI
Where to start
A small first engagement with a verifiable outcome
This reduces risk before assigning a critical module or an entire project.
1–2 weeks
Performance cycle
Reproducible measurement, profiling and first changes.
- baseline benchmark
- hot-path analysis
- prioritized optimization plan
- before/after measurements
2–4 weeks
Technical hypothesis validation
A working prototype of the riskiest part of the solution.
- bounded prototype
- performance or compatibility test
- documented limitations
- full implementation estimate
duration after triage
Module stabilization
Crashes, races, memory corruption or a release-blocking defect.
- reproducible scenario
- root-cause analysis
- fix and regression test
- safe-release plan
Minimum paid engagement: 3 business days, from RUB 200,000 or the equivalent in another currency. After a short introductory call, scope, deliverables, validation method and price are fixed.
Engineering services
Critical components for infrastructure products
Drivers, high-performance networking, storage and data protection — dedicated development areas with concrete problems and project results.
01 · TELECOM
XDP/eBPF, DPDK and telecom data planes
GTP, DPI, UPF/PGW, CGNAT, NUMA, NIC offload and measurable performance.
Open direction →02 · KERNELDrivers and system filters
Linux, Windows and macOS device, block, file and endpoint components.
Open direction →03 · DATAStorage and OS-level data protection
BIO, RAID/EC, snapshots, encryption, access control and recovery.
Open direction →DRIVER DEVELOPMENT
Driver development: specialized services
Dedicated pages for distinct engineering needs, from a new device to a rare system failure.
Windows driver development
WDM, KMDF, UMDF, PCIe, USB, DMA, PnP, and release preparation.
→Linux drivers and kernel modules
Kernel modules, devices, block and network paths, and kernel-version maintenance.
→Windows file, storage, and network filters
Minifilters, storage filters, WFP, and system-level policy enforcement.
→BSOD and driver debugging
Crash dumps, Driver Verifier, WinDbg, races, memory corruption, and lost I/O.
→FREE TOOLS
Tools for networking, storage and technical assessment
PPS, GTP-U and XDP calculations, NIC/NUMA/IRQ analysis, FIO benchmark sanity checking and startup technical assessment. Data is processed locally in the browser.
Technical advisory
Independent technical judgment before an investment, acquisition or major product transition
For investors, boards and founders: what R&D can really deliver, where deal risks sit and which technologies should be protected.
INVESTMENT / M&A
R&D due diligence and startup rating
Team, product, architecture, delivery, QA, support, innovation and IP tested through evidence and confidential interviews.
Review the method → FUNDRAISING / EXITIP readiness for funding or M&A
Code ownership, licensing, core-technology map, patent coverage, protection gaps and technical data-room readiness.
Review readiness →FREE LOCAL AI LAB
Open Local RAG Lab →Build a RAG pipeline and inspect every stage
Environment setup, documents, chunking, retrieval, the augmented prompt, local generation through WebGPU or CPU/WebAssembly, and source-support checks—all visible and browser-local.
FREE TOOL
Open self-assessment →Quick startup self-assessment
25 questions, AAA–C rating, spider chart and HTML report.
Four confidential reports for a venture fund
Carrier-grade network virtualization, a BI platform, AI developer tooling and an IT-infrastructure digital twin. Each report covered rating, red flags, product, architecture, engineering quality, innovation and recommendations.
Review the anonymized track record →Verified CloudLinux outcome
Within three months the team learned to own the patent process, filed several applications and built a pipeline of additional ideas.
Open recommendation ↗Project examples
How similar problems were addressed
Confidential customer information is omitted; test configuration and methodology can be discussed technically.
Telecom project · Linux/XDP · GTP-U
From 16 to 400 Gbit/s: two ConnectX-6 adapters across two NUMA nodes
Starting point
Initially, the mobile-core user plane processed about 16 Gbit/s through one port of a dual-port Mellanox ConnectX-6 adapter. The task was to optimize the software data path and scale processing across all network interfaces in the server.
Contribution
GTP-U data-path optimization included per-CPU state, NUMA-local memory, RX/TX queue affinity, cache-line separation, batching, lock-free exchange and NIC offload. The second port on the first adapter was then enabled, and a second dual-port ConnectX-6 was installed on the other NUMA node, with memory, queues and workers kept local to each adapter.
Result
16 → 400 Gbit/s aggregate
Validation
One server and the same GTP-U traffic profile; validation through application counters, per-port NIC counters, CPU and NUMA load distribution, and repeated runs. The final configuration used four 100GbE ports across two adapters.
Storage vendor · Linux
A custom block and file layer
Starting point
Existing components did not fit the data model, resilience requirements or performance target.
Contribution
Virtual block devices, BIO mapping, RAID/EC, metadata, thin provisioning, copy-on-write snapshots, SSD/NVMe write path, recovery and rebuild flows.
Result
Architecture and critical low-level components for distributed storage.
Validation
Failure recovery, metadata integrity, degraded reads and writes, throughput and latency measurements.
Data-protection product · 3 OSes
One systems layer across Linux, Windows and macOS
Starting point
Three platform implementations diverged in behavior, interfaces and release timing.
Contribution
Shared C++ core, thin OS adapters, common memory, synchronization, I/O and event abstractions, versioned interfaces and a common test matrix.
Result
One architecture for drivers, kernel modules and system services with localized platform differences.
Validation
Automated builds and regression tests across platforms, interface compatibility and upgrade scenarios.
Endpoint / security
Data control below the application layer
Starting point
Protection and access policies had to apply independently of application and user behavior.
Contribution
File and disk filters, policy cache, transparent encryption, asynchronous auditing, bypass protection and fail-safe behavior.
Result
OS-level policy enforcement and auditing integrated into an existing product.
Validation
Negative tests, bypass attempts, dependent-service failures, audit coverage and data-integrity checks.
PUBLICATIONS
Technical articles grounded in real systems engineering
Driver, legacy-code, storage and engineering-method analyses — not generic documentation summaries.
Professional background
Professional track record and recommendations
The professional background can be checked through public work, patents, profiles and client recommendations.
14 years at Acronis
Written recommendations from Acronis executives
Names are withheld; roles and the substance of the recommendations are preserved.
Acronis and Acronis Labs
Technical Fellow, systems engineering and leadership of an R&D initiative focused on storage and data protection.
Open publication ↗ Client recommendationCloudLinux
Within three months the team learned the patent process, filed several applications and created a pipeline of further ideas.
Open recommendation ↗ Patents and citations60+ patents
More than 2,400 citations and an h-index of 32 on the author's Google Scholar profile.
Open profile ↗Public work
Code and publications
GitHub, systems and R&D articles, an early technical footprint in OSR.
Engineering leadership and product creation
Senior Vice President, Acronis: Maxim was described as essential to the company’s growth from a small startup into an international software vendor, while remaining a hands-on product contributor and leading hundreds of engineers.
Chief Strategy and Operations Officer, Acronis: their work included international R&D-center strategy and innovation priorities for sustainable company growth.
Project outcome
What the customer receives
Deliverables and acceptance criteria are defined before the stage begins.
Technical assessment
Constraints, risks, bottlenecks and a justified scope.
Architecture
Components, interfaces, data path, concurrency model and failure modes.
Prototype and measurements
Validation of the riskiest assumption before full implementation.
Release-ready code
Source, review, profiling and integration into the customer repository.
Tests and documentation
Regression tests, build instructions, architecture decisions and knowledge transfer.
Release support
Integration, lab diagnostics, stabilization and further optimization.
NDA before code disclosureConfidential material can be discussed after an agreement is signed.
Customer-owned resultSource code and agreed IP rights are transferred under the contract.
Work in your environmentCustomer repository, CI, tracker, lab and engineering rules.
Milestone acceptanceVerifiable criteria instead of an undefined purchase of engineering hours.
Engagement stages
How the work is structured
A large project starts only after the critical assumptions are validated.
- 01
Problem summary and introduction
Platform, current implementation, failure point, target load and acceptance criterion.
- 02
NDA and access to material
Code, diagrams, profiles and measurements when required.
- 03
Assessment or bounded first stage
Validate the highest-risk area with an agreed outcome.
- 04
Full implementation
Code, tests, profiling, documentation and integration in milestones.
- 05
Handover and release support
Knowledge transfer, production launch and issue resolution.
Accountability
Maxim Lyadvinsky
Systems architect and C/C++ kernel engineer
How KernelSoft is structured
This is an independent expert practice. I personally own the architecture, critical code, diagnostics and final technical review. For larger scopes, additional engineers or domain specialists are involved only after roles, access and accountability are agreed with the customer.
More than 28 years in systems engineering, including 14 years at Acronis. Work has covered drivers, storage, information security, virtualization, high-speed packet processing and cross-platform systems infrastructure.
Co-author of more than 60 patents. Led Acronis Labs, an R&D initiative focused on storage, performance and data-protection technologies.
28+years of experience
60+patents
2,400+citations
32h-index
Discuss the project
Describe the problem, current implementation and target outcome
Include the platform, what already exists, where the problem occurs, the target load and the acceptance criterion. This is enough to choose a sensible first stage and validation method.
Target response time: one business dayNext: a short introduction → an NDA if needed → agreement on the first stage.