# Routine Automation at hh.ru — Full Technical Details

> Portfolio site showcasing two internal automation tools built by Pavel Sidorov, Traffic Manager at hh.ru (HeadHunter — Russia's largest job platform, ~50M monthly users).

## Geo Synchronizer

### Problem
hh.ru operates 9 independent services that each maintain geographic data (cities, regions, metro stations). When geographic changes occur (new cities, renamed regions, merged territories), a traffic manager must manually update each service — a process taking ~40 minutes per change, prone to inconsistencies.

### Solution
A Python-based synchronization engine that propagates geographic changes across all 9 services through a single interface.

### Architecture
- **Backend**: Python, FastAPI REST API with 4 endpoints
- **Frontend**: Streamlit interactive dashboard
- **Deployment**: Railway (API), Streamlit Cloud (UI)
- **Codebase**: 9,972 lines of Python, 73 automated tests

### Algorithm (5-step sync)
1. **Collect** — Pull current geo data from all 9 services via APIs
2. **Normalize** — Standardize naming conventions, encoding, ID formats
3. **Compare** — Diff detection across services, identifying mismatches
4. **Resolve** — Apply conflict resolution rules (4 sync modes: full, partial, append, validate-only)
5. **Propagate** — Push resolved data to all services, with rollback on failure

### Sync Modes
- **Full Sync**: Complete replacement of target data
- **Partial Sync**: Update only changed fields
- **Append**: Add new entries without modifying existing
- **Validate Only**: Dry run — report differences without changes

### Key Metrics
- Sync time: 40 min → 1.5 min (96% reduction)
- Coverage: 18,000+ cities and regions
- Annual savings: 660 hours/year
- Error rate: Near-zero (automated validation vs manual copy-paste)

### Links
- App: https://geosynchronize.streamlit.app/
- API: https://geo-sync.up.railway.app/docs

---

## SMS & Email Generator

### Problem
Creating personalized SMS and email campaigns for different Russian regions requires manual text composition with correct morphological forms (Russian language has 6 grammatical cases). A traffic manager spends ~15 minutes per campaign crafting messages with proper declensions.

### Solution
A morphology-aware text generation engine that produces grammatically correct campaign messages from templates.

### Architecture
- **Backend**: Python, FastAPI REST API with 4 endpoints
- **Morphology Engine**: 707 lines of Python, handles Russian noun/adjective declension
- **Frontend**: Streamlit interactive dashboard
- **Templates**: 25 pre-built campaign templates with variable slots
- **Deployment**: Railway (API), Streamlit Cloud (UI)

### How It Works
1. User selects a campaign template and target regions
2. Engine extracts geographic names and determines grammatical gender/number
3. Morphology module declines names into required grammatical cases (genitive, prepositional, etc.)
4. Generated messages are previewed and exported (CSV/clipboard) for bulk sending

### Key Metrics
- Generation time: 15 min → 1 min (93% reduction)
- Templates: 25 campaign types
- Annual savings: 130 hours/year
- Grammatical accuracy: >99% (vs ~90% manual)

### Links
- App: https://smshhru.streamlit.app/
- API: https://smshh.up.railway.app/docs

---

## Combined Impact

| Metric | Before | After | Savings |
|--------|--------|-------|---------|
| Geo sync per task | 40 min | 1.5 min | 96% |
| SMS/Email per campaign | 15 min | 1 min | 93% |
| Total annual hours saved | — | — | 790 hours |
| Error reduction | Manual | Automated | ~99% |

---

## Technology Stack

- **Site**: Next.js 14, React 18, Remotion (video framework), Tailwind CSS, TypeScript
- **Tools Backend**: Python, FastAPI, Streamlit
- **Hosting**: Vercel (site), Railway (APIs), Streamlit Cloud (dashboards)
- **Database**: Neon PostgreSQL (feedback collection)

## Author

**Pavel Sidorov**
- Role: Traffic Manager at hh.ru
- Email: p.sidorov@hh.ru
- Site: https://sidorov-vibecode.com