Data & OSINT Engineer · Web Scraping · Backend Development
I build tools and workflows for collecting, processing, and verifying public information: web crawlers, data pipelines, backend APIs, and structured research datasets.
My background combines 20+ years in software engineering with 5+ years focused on OSINT and investigative research.
Core stack: Python, SQL, PostgreSQL, JavaScript/TypeScript, Playwright, Docker, Linux.
I use AI for extraction, classification, coding, and workflow automation, combined with deterministic checks and source-based verification.
Selected projects · Upwork · LinkedIn · Substack · Email: m.marshak@proton.me
Workflow: source collection → parsing and extraction → normalization and enrichment → validation and deduplication → structured data → research and analysis.
A modular Python-based platform for collecting heterogeneous web data with Playwright, Puppeteer, Camoufox, HTTP requests, and browser-profile environments. Collection scenarios are separated from individual sources to make workflows reusable and configurable.
Engineering focus: parallel and distributed collection, browser automation, reusable processing logic, and AI-assisted scenario generation. Public case study: the UFO archive collection workflow below. Technical walkthrough available on request.
A structured snapshot of the U.S. Department of War UFO archive, collected with Oriol. The workflow transforms heterogeneous public source material into JSON for querying, archival research, enrichment, and downstream processing.
Input: public web archive. Work: crawling, extraction, normalization, and organization. Output: a reusable machine-readable research snapshot.
Read the archive collection case study — public workflow documentation; downloadable JSON is not currently included.
A repository-metadata collection covering OSINT, data collection, automation, infrastructure, and security research. The 14 November 2026 full snapshot contains 4,484 repositories; the public repository provides a 100-record sample and the Python generator.
The generator collects GitHub metadata and produces JSON, Markdown, PDF, topic lists, and start.me bookmarks. Dated snapshots support comparison of newly added repositories.
Generator and public sample · Full release and access details
A Chrome extension and TypeScript/Fastify backend that collect domain and page signals through RDAP, DNS, TLS, and metadata checks. The Docker Compose stack includes PostgreSQL-backed caching and shared TypeScript contracts.
This is an MVP with explainable rules, with limitations documented in the repository. Signals support manual source verification; they do not establish that a site is truthful, false, or malicious.
Source Credibility Overlay: code, setup, and limitations
A toolkit for acquisition and processing of the U.S. DOJ Epstein Archives: automated downloading, PDF parsing, duplicate detection, redaction analysis, and extraction for investigative review. The project processed approximately 89 GB of source material.
Private repository · Technical walkthrough available on request. Processing figures and interpretation of redaction-analysis outputs can be discussed with the relevant snapshot and methodology.
A workflow for transforming heterogeneous professional-profile data into portable Markdown, JSON, PDF, and related research records. It was used while reorganizing a collection that grew to approximately 50,000 profiles.
The project demonstrates format conversion, entity-centric organization, searchable records, and integration with Markdown/Obsidian workflows. It is an experimental workflow with documented limitations.
linkedin2md: workflow and examples
| Area | Tools and methods |
|---|---|
| Collection | Python, Playwright, Puppeteer, Camoufox, HTTP APIs, scraping and crawling |
| Data | SQL, PostgreSQL, ClickHouse, JSON/CSV, normalization, enrichment, deduplication |
| Backend | TypeScript/JavaScript, Node.js, Django, REST APIs, Celery, RabbitMQ |
| Automation | Bash, Ruby, Perl, Docker, Linux, AWS |
| AI-assisted workflows | LLM extraction and classification, coding agents, source-based verification |
My broader engineering background includes legacy systems, integrations, databases, infrastructure, and software in C/C++, Java/.NET, PHP, and other stacks.
OSINTech’s Timeline is a weekly collection of OSINT tools, services, datasets, investigations, and research methods, maintained since 2023. It combines source monitoring, manual review, classification, and publication.
Timeline archive on GitHub · OSINTech’s OSINT Anatomy on Substack · Start Here
Evidentia / Sourcely: research and engineering around source verification, evidence organization, data processing, and AI-assisted investigative workflows.
Tesari.AI: OSINT engineering and research-data curation, including collection, source coverage, enrichment, data-quality assessment, and QA of generated investigative reports.
- just-ai: my extension of the upstream
justcommand runner, adding project-context analysis, local command-risk checks, and optional AI assistance. Upstream attribution and fork details are documented in the repository. - fcc-model-check: diagnostics for model availability and routing between FCC and NVIDIA NIM.
- Scam Surface Mapper: a modified fork for mapping web infrastructure in OSINT research.
- IP & Location Widget: a lightweight widget for start.me boards.
- Martin: an experimental animated companion for Codex Desktop.
The project history includes further prototypes, evidence-capture work, publications, and earlier engineering projects.
I work openly under my real identity and support independent research, investigative journalism, human-rights work, and democratic civic initiatives.
I publicly oppose Russia's war against Ukraine and support projects documenting state and war crimes.
I do not collaborate with Russian public-sector institutions, Russian banks, or import-substitution-related companies.
For data collection, backend development, OSINT tooling, and research projects: Upwork, LinkedIn, or m.marshak@proton.me.
If my tools or publications help your work, you can support OSINTech on Buy Me a Coffee.

