EU AI Act Scanner · Impending launch
In final QA
Problem. Operators of high-risk AI systems face Regulation 2024/1689 with a hard deadline of 2 Dec 2026 for Article 50(2) synthetic-content labeling, and 2 Dec 2027 for Article 72 Post-Market Monitoring. Most companies have no inventory of their AI systems, no per-system classification, and no audit-ready documentation.
Approach. Rule-based 5-tier risk classifier mapping purpose + affected populations + data types to Articles 5 + Annex III. Role-branched questionnaire (provider / deployer / importer / distributor) scoring 8 obligation areas. Generation of 4 document types: Technical Report, Annex V Declaration, FRIA, Post-Market Monitoring Plan. WORM-anchored audit trail.
Outcome. From zero classification to audit-ready documents in under 2 hours per system. Jurisdictional language support per Article 47(2). Built to be the tooling layer between operators and the notifier body.
TuneCV · In development
Backend 100% · Frontend 60%
Problem. Job seekers rewrite their CV from scratch for every application. Format friction eats hours. ATS systems reject well-qualified candidates for phrasing issues, not skill gaps.
Approach. Multi-format parser (PDF / DOCX / TXT) with intelligent text extraction. ATS scoring against any job description. GPT-4-driven per-section rewrite suggestions with measurable ROI (keyword coverage, action verbs, quantified achievements). Kanban application tracker. Polar.sh + Stripe billing integration.
Outcome. 390+ tests across backend and frontend. Feature gating system done. Polar webhook handling live. Frontend checkout UI functional. AI feature UIs remaining.
CheckCV · In development
Greenfield + FastAPI + Next.js
Problem. Recruiters handle high-volume CV intake. Format normalization and rubric scoring eat hours per week. Per-organization scoring rubrics drift between reviewers.
Approach. Multi-format parsing with AI scoring. Custom scoring rubric per organization with tier-aware weights and audit trail. Job matching with compatibility scoring. White-label branding per buyer. 7-locale i18n via next-intl. Tauri 2 desktop shell for offline review.
Outcome. Architecture pinned. Stack selected (FastAPI + Next.js 16 + PostgreSQL 16 + Redis + Celery). Supabase Auth. Polar.sh MoR. Ship-readiness audit in flight.
PropVector AI · In development
OSS-public + commercial layered
Problem. Real-estate search runs on filters and keywords. Brokers and PropTech platforms need semantic search that understands "quiet neighborhood with good schools" or "investment property with high ROI potential" — not just price + beds + baths.
Approach. Dual strategy. Public open-source (OpenStore) ships the core architecture: ChromaDB vector search, MMR reranking, 13 LLM providers, 9 languages, financial tooling, multi-year price forecast. Commercial tier (PropVector AI) layers live MLS feeds, accounts, favorites, saved searches, AI CRM, enterprise-grade SLA.
Outcome. Public OSS at 285 GitHub stars. Commercial build on VPS5 with full Next.js + FastAPI + ChromaDB + PostgreSQL + Redis stack. Polar.sh MoR for paid tiers.
