The problem
Job outreach is usually split between Google Maps, spreadsheets, contact lists, CV files, and repeated prompts. The result is slow research, duplicate data, and very little control over personal information.
Back to selected work
01Project mission dossier
Intelligent workflow system · 2026
ApplyFlow AI
A private lead research and personalized outreach workspace that turns company discovery, contact organization, CV context, and multilingual messaging into one reviewed workflow.
- Role
- Product designer & full-stack developer
- Status
- Working local-first MVP
Project snapshot
- Role
- Product designer & full-stack developer
- Status
- Working local-first MVP
- Platform
- Responsive web application
- Deployment
- Local SQLite workspace
- Scope
- Research, CRM, CV context and reviewed outreach
- Repository
- moaadahtchaou/ApplyFlow-AI
The solution
I designed one local workspace that researches companies, reviews and deduplicates leads, organizes contacts, extracts reusable candidate context, and generates editable outreach messages without sending anything automatically.
Technical system
Verified components and boundaries found in the project source and documentation.
- 01Lead research
Cancellable Playwright searches
- 02Review inbox
Canonical identity and deduplication
- 03Local CRM
Companies, contacts and SQLite
- 04Candidate context
Reviewed PDF or text CV extraction
- 05Message generation
Optional NVIDIA AI with local fallback
Local-firstprivacy model
4outreach languages
End-to-endproduct ownership
What it does
Google Maps research with progress and cancellation
Multi-signal lead deduplication
CV extraction and structured candidate context
Editable multilingual AI messages
Product walkthrough02
Engineering decisions
Useful results, not repeats
Nearby Google Maps searches frequently surface the same highly ranked companies.
- Decision
- Combine Maps URLs, domains, phones, coordinates, and normalized names into canonical identities.
- Engineering effect
- Multi-city runs can skip known entities and continue toward new reviewed leads.
AI that stays honest
CV context can become unreliable if generation silently fills missing details.
- Decision
- Require reviewable candidate context and prompts that forbid invented experience, achievements, or contacts.
- Engineering effect
- The generated draft stays editable and grounded in user-approved facts.
Ownership map
Product designer & full-stack developer
- 01
Defined the product scope and complete workflow
- 02
Designed the Prisma data model and migrations
- 03
Built the Next.js interface, automation, AI integration, validation, and tests
Outcome
The MVP demonstrates a complete privacy-conscious product workflow: browser automation, data modeling, resilient AI fallbacks, responsive product design, validation, and testable business rules.
Riwayat
Arabic publishing platform