Software for legal research, language learning, and public services.

My work covers source ingestion, normalized data models, authentication, rate limiting, APIs, collaborative interfaces, and deployment.

See the systems ↓CV / roles / publications

ESTATE / 01CASE LAW EXPLOREROPERATED

Case Law Explorer platform

Research infrastructure

Rechtspraak, HUDOC, and CELLAR records enter one legal-data schema, then reach the research surface through a bounded API and server-side proxy.

Python · Airflow · PostgreSQL · SvelteKit · Docker

NLRechtspraakXML

ECtHRHUDOCAPI

EUCELLARSPARQL

extract / normalizeAirflow + ECLI

unified data planecle_v2cases · corpus records · text · citations

15 JSON routescitations-apisearch · text · expand · methods

optional policy pathKeycloak / OIDCcaselaw-accesssubject · scope · cost · quota

01SvelteKit proxy + UI

02Coolify estatemigrate · backup · SQL · ETL

Authentication is optional. When enabled, Keycloak identifies the caller and caselaw-access separately decides scope, route cost, and quota.
/api/combined searches all three corpora; /api/combined/expand replaces the retired generic network route.
Coolify composes the UI, API, database, migration, backup, SQL maintenance, and Airflow services; Airflow keeps a separate bookkeeping database.

Case Law Explorer architecture

Hayeren

Language infrastructure

A shared Armenian domain model connects morphology, study scheduling, the lexicon, collaborative notes, and every learning surface.

React · Vercel Functions · Neon · Liveblocks/Yjs

first-party auth proxyNeon Auth / JWKS

Reader

Course

Dictionary

Classrooms

portable domain@hayeren/coremorphology · SRS · mastery · permissions

@hayeren/api-client

generated content + tokens

durable stateNeon Postgreslexicon · events · notes

Liveblocks / Yjsshared notes + rooms

Surface forms resolve to stable lemmas with lect, provenance, senses, and valency.
Reviewed Markdown generates the content package; design values generate the token package.
Learner state and note projections live in Neon; an active shared Y.Doc lives in Liveblocks, and private pages open no room.

Hayeren architecture

Commonfold

Private knowledge system

One input produces a reviewed capture or a source-linked answer. Postgres remains canonical while collaboration, version history, retrieval indexes, and portable Markdown project from the same note.

Next.js · PostgreSQL/pgvector · Claude · TipTap/Yjs · Clerk

one entry pointCapture / queryintent classification

canonical archivePostgresnotes + Markdown projection

Entity resolutionstable people · places · concepts

pgvectorembeddings + retrieval

Claude toolssources exposed

TipTap / Yjscollaborative Markdown

Version historyactor + instruction + restore

Postgres is the source of truth; Markdown remains the portable representation.
Capture exposes create/update, type, folder, and unresolved-entity decisions before write.
Keyword and semantic evidence stays linked through answer generation.

Commonfold architecture

Mobility in Change

Client experience platform

Four localized routes take an industrial client from a cross-border need to one of two consulting practices and a qualified inquiry for Christine.

4 locales · 2 practices · 3 inquiry steps · live client editing

ITpublic route

ENpublic route

DEpublic route

FRpublic route

first impressionAmbient filmposter on reduced motion or save-data

decisionTwo practicescommercial · sustainability

qualified path3-step inquiry

human responseChristine reviewsemail · clarify · private meeting

route preserved across localeconsent before analyticsclient edits live preview

The same visual hierarchy and service path is authored in Italian, English, German, and French.
One ambient film establishes tone; a poster protects reduced-motion and data-saving visits.
The latest maintained revision asks for need and readiness before contact details; production promotion is still pending.

Mobility in Change architecture

Ask about David’s work.

Test a role, research question, or project brief against the public record—from current systems to education and earlier work.

Would David be a good fit for a legal research software team?What has David built with retrieval and AI?Which work demonstrates full-stack architecture?

Open the conversation

Market vendor in Yerevan — Yerevan · Photography archive

Education, languages, photography, and earlier work.

Computer science at Maastricht University; Italian, English C2, German B1, Arabic A2, and Eastern Armenian in progress. The archive includes 65 photographs and detailed records from school projects, climate organising, teaching, publications, and early software.

Index

Pages

Research & publications

Projects

Fieldwork & access

Elsewhere