Den Patin

Product Engineer

VOIS — AI vocal coach for iOS · contract, remote

• Full ownership of the backend and infrastructure: API, background audio processing, admin tooling, deployments, monitoring.
• Building the live-feedback layer: a compact model trained on per-note scores and probabilities from the app's signal-processing core, so feedback follows a vocal-pedagogy methodology.

Backend Engineer

StagedByAI — AI interior staging · contract, remote

• Built the entire backend and infrastructure from scratch, image-generation pipeline included, and ran it through two years of growth.

Lead Developer / Backend Engineer

nebenan.de (Germany's largest neighborhood network), shyftplan (workforce scheduling), Rentumo (rental listings) · contract, remote

• Backend and architecture across three unrelated domains; led cross-functional teams of up to 8 engineers.
• Hosting costs −30% and delivery speed ×2: legacy monoliths migrated to Ruby 3+ / Rails 7–8, CI/CD modernized after a cost analysis.
• Critical query times −90% in a multi-terabyte PostgreSQL database; incident resolution −40% through monitoring and profiling.

Senior Software Engineer / Tech Lead

Deutsche Telekom IT Solutions (T-Systems) · Saint Petersburg

• 12 internal products, 9 of them built end-to-end (all under NDA); tech lead of teams up to 8.
• Designed and owned the company-wide employee master-data service: SAP, 1C and Active Directory reconciled into one cleared source for 10+ downstream systems of a 2,500-employee organization.
• Tech lead of the system documenting the operational state of T-Mobile Germany's national mobile network (GSM/GPRS/UMTS/LTE): country-scale infrastructure data with geospatial views.
• Most-loaded dashboard 10× faster (3 s → 300 ms) for 2,500+ concurrent users: indexing, stored procedures (utPLSQL), query redesign.
• Turned recurring bureaucracy into software: workforce-cost forecasting, indoor navigation and seating/cost planning, national facilities web-GIS, company-wide event booking.

Lecturer (part-time)

ITMO University; Peter the Great St. Petersburg Polytechnic University

• Taught Linux, Ruby and web development with Ruby on Rails (classes up to 20); mentored 16 interns, 60% of whom moved into full-time engineering roles.

Research Assistant

Philological Research Institute, Saint Petersburg State University

• Grant “Quantitative linguistic parameters for definition of stylistic features and domains of text”: corpus statistics for register identification.

Programmer

Other Accents Studio · Saint Petersburg

• Web applications and the studio's sites: PHP (CodeIgniter), JavaScript, jQuery.

Technical Writer / Translator

Nexenta Systems; Systematica; Odnoklassniki · Saint Petersburg

• Product documentation, layout and localization; technical translation and in-country review (English, Japanese, German).

TaiwanCards

Corpus-linguistics and pronunciation-scoring platform for Taiwanese Mandarin; sole author of corpus, models and scorer

• Segmentation: Viterbi decoding over a Kneser–Ney bigram model, EM re-segmentation, relative-entropy pruning; evaluated on a 510-item gold set with McNemar's test and a paired bootstrap on F1.
• Lexical relations: PPMI with context-distribution smoothing and logDice; dense SVD embeddings benchmarked on the same gold set and rejected.
• Pronunciation scoring from acoustics (F0 contour, VOT, formant trajectories), server-side, no cloud ASR; FSRS-6 scheduling.
• Source: study app (Apache-2.0), corpus pipelines.

dsprb and dtwrb

Open-source Ruby libraries for speech signal processing

• dsprb: speech DSP in pure Ruby — radix-2 FFT, mel filterbanks, MFCC, YIN F0 estimation, LPC formants (source).
• dtwrb: dynamic time warping with a Sakoe–Chiba band, DBA averaging for variable-length sequences (source).

taiwan-corpora

Three open datasets for Taiwanese Mandarin on Hugging Face

• twngrams (CC0), twsyllables (CC BY 4.0), twfilter-tables.
• Pipelines (MIT, GitHub): twpipeline, nine JSONL stages from web crawl to n-gram tables; twfilter, provenance analysis — script, orthography, lexical origin, punctuation.

DenMed

Local-only pipeline for multilingual medical records; in progress

• Lab PDFs and scans in any language → LOINC codes and SI units → longitudinal charts against multi-source reference ranges; local models only.

Research prototypes

• Cross-spectral analysis of arterial pressure and cerebral blood-flow velocity — coherence, B/M-wave phase shift, autoregulation indices — for the Polenov Neurosurgical Institute, 2020.
• Search relevance and bidialectal ranking for ironau.ru, the primary online dictionary of Ossetic, 2018–2019.
• Kyrgyz morphological analyzer in Prolog, 2017.

M.Sc. in Applied Informatics (Computational Linguistics)

Saint Petersburg State University

Thesis: corpus-scale automation of Greenbergian quantitative typology — morphological indices (synthesis, agglutination, suffixation) and register-discriminative statistics over 1–12M-token corpora of Hungarian, Japanese and Turkish stratified by functional style, where prior work from Greenberg (1960) to Kasevich (2009) relied on manual samples of 100–1,000 syllables. Result: a typological profile is a property of a register, not only of a language; validated against published manual counts.

B.Sc. in Informatics and Computer Science

ITMO University

Thesis: automatic interlinear glossing (Leipzig Glossing Rules) of Hungarian, Turkish and a constructed polysynthetic language used as a controlled test case, its gold segmentation known by construction; rule-based morphological segmentation constrained by vowel harmony and morphotactics; a hand-implemented feedforward network (backpropagation) disambiguated homonymous affixes. Delphi, 2013 — before ML frameworks made this routine.