Autonomous AI platform for wholesale intelligence. Product enrichment, competitor monitoring, demand forecasting, and sentiment analysis — powered by 8+ data sources and self-healing infrastructure.
Horus layers intelligence on top of your existing infrastructure — no rip-and-replace, no disruption, no six-month implementation.
AI-driven demand forecasting, automated procurement triggers, and smart inventory rebalancing. Your ERP finally thinks ahead instead of just recording the past.
Route optimization, warehouse automation, and real-time shipment intelligence. Reduce logistics costs by up to 23% with ML-powered operational insights.
Customer behavior analysis, churn prediction, and hyper-personalized engagement. Know what your customers need before they do — then act on it automatically.
From connection to continuous evolution — Horus integrates seamlessly with your existing stack in days, not months.
Plug into your existing ERP, WMS, CRM, and databases via secure REST APIs. No data migration, no downtime.
AI scans your data patterns, workflows, and anomalies — building a real-time model of your operations.
Automated recommendations, smart alerts, and self-healing actions — all with your approval before execution.
Continuous learning from every outcome. Horus compounds intelligence — month over month, year over year.
Each module is live, JWT-protected, and purpose-built for food wholesale operations. Click any card to launch.
Six live intelligence modules, a production-hardened microservices stack, autonomous self-healing infrastructure, and role-based user management — all running in production today. Every technology decision is documented below.
Sentinel is not a monitoring tool that queues alerts for humans to clear. It is a fully autonomous agent that detects problems and resolves them without human intervention — continuously, 24 hours a day, 7 days a week, with zero manual maintenance required. When a service fails, Sentinel restarts it. When code quality degrades, Sentinel detects every issue, applies fixes, and re-scans to confirm the corrections took effect. When disk fills, Sentinel prunes Docker artifacts and rotates oversized log files. When a problem recurs three times, Sentinel escalates with a data-driven recommendation for a permanent architectural fix. Nothing sits in an alert queue — Sentinel closes every loop itself.
Sentinel learns continuously from its own operational history. Every failure, every healing action, every code finding, and every outcome is stored and indexed in PostgreSQL. Deduplication by MD5 hash tracks which problems are systemic vs. one-off. MTTR analytics measure how effective each fix type is over time. Weekly pattern reports surface the top recurring incident types from the past week — a self-generated, prioritised improvement backlog produced without any human input. The longer Sentinel runs, the more precisely it identifies structural weaknesses before they cause outages.
Code correction is a first-class, fully automated capability. Every 6 hours Sentinel runs Ruff static analysis plus 15 custom detectors across the entire Python codebase — covering hardcoded secrets, SQL injection via f-strings, shell injection (subprocess shell=True), N+1 query loops, time.sleep() in async context, bare except clauses, and eval()/os.system() calls. Safe rules (F401, F841, E711, E712, W291–W293) are fixed immediately and automatically with no developer action required. The /refactor pipeline goes further: it applies a full ruff check --fix pass plus ruff format across the entire watched codebase, then restarts the backend to apply the changes — a complete code quality improvement cycle that Sentinel runs autonomously or on operator demand via Telegram.
Technical: standalone FastAPI microservice on port 9000, five APScheduler jobs (AsyncIOScheduler, Europe/Zagreb). Fast loop 30 s: API health + response latency, container CPU/memory/restarts, host CPU/RAM/disk/load. Slow loop 5 min: PostgreSQL connections + DB size, Nginx 5xx spike detection, SSL certificate expiry, Product Intelligence latency. Telegram bot with 11 commands, inline Approve/Deny keyboards (5-min auto-approve), warning digest every 30 min, daily 08:00 CET report, weekly Monday 09:00 CET summary. 17 Prometheus metrics on isolated custom registry at /sentinel/metrics.
Cascading data enrichment pipeline querying OpenFoodFacts (multi-region IT/NL/EN), UPC ItemDB, Barcode Lookup, Icecat, SerpAPI+Gemini AI, Google CSE, Rainforest/Amazon across IT/NL/DE markets. Field-level source attribution tracking. Confidence scoring (high/medium/low/minimal) based on source reliability and data completeness. Smart merge with first-non-empty precedence. 24h TTL caching layer. Image proxy for CORS-free delivery.
Automated price tracking across Dutch retail: Albert Heijn, Jumbo, Crisp, Picnic. Historical price trend analysis with change detection alerts. Cross-competitor product matching by EAN. Market positioning insights with price percentile ranking. Scheduled scraping with respectful rate limiting and User-Agent rotation.
Scikit-learn regression models with seasonal decomposition for Italian food product demand prediction. Category-specific seasonal patterns: Christmas panettone spikes, summer BBQ trends, Easter colomba cycles. Confidence intervals on predictions. Historical sales data integration with ERP systems. Retraining pipeline with new data ingestion.
Dual-engine sentiment analysis combining TextBlob polarity with VADER compound scoring. Multilingual support: English, Italian, Dutch, Croatian. Google Reviews and Trustpilot aggregation for brand monitoring. Competitor sentiment benchmarking. Temporal trend analysis for reputation tracking. Configurable alert thresholds for sentiment drops.
TheMealDB integration with 300,000+ recipe database. Anthropic Claude AI recipe generation from ingredient lists. Product catalog cross-linking: recipe ingredients mapped to 3,000+ SKU catalog. Wine pairing suggestions. Dietary tag classification: vegan, gluten-free, halal, kosher. Multi-language recipe content.
Containerized microservices on Docker Compose with resource limits (CPU/RAM per service). PostgreSQL 16 with async SQLAlchemy and connection pooling. Nginx reverse proxy with SSL/TLS (Let's Encrypt), HSTS, CSP, X-Frame-Options DENY. UFW firewall with minimal port exposure. JWT HS256 authentication. Environment-based secrets management (zero hardcoded credentials). Grafana + Prometheus observability stack with custom dashboards.
FastAPI with automatic OpenAPI/Swagger documentation auto-generated from Pydantic v2 schemas. Async/await throughout every endpoint for fully non-blocking I/O — no thread pool bottlenecks under load. Request/response validation with detailed error messages. CORS-configured for multi-origin access. Per-endpoint rate limiting. Dedicated health check endpoint for container orchestrator integration. JWT Bearer middleware applied selectively — public endpoints remain open while intelligence modules require auth. Full interactive API docs at /api/docs with Try It Out support for every endpoint.
JWT HS256 tokens carrying user ID, email, and admin-role claim — verified on every protected request via FastAPI dependency injection. bcrypt password hashing (12 rounds). HSTS with 1-year max-age enforced at Nginx. Content-Security-Policy with explicit source allowlisting: script-src restricted to self and unpkg CDN, frame-ancestors none to prevent clickjacking. UFW firewall — only ports 80 and 443 exposed publicly, all internal services communicate on Docker bridge network. Zero hardcoded credentials: all secrets injected via Docker environment variables at runtime. GDPR-compliant account self-deletion endpoint purges all personal data. Role-based access control enforced at two layers: JWT claim checked client-side for UI gating, PostgreSQL column re-verified server-side on every admin action.
Full user lifecycle: registration with email, phone, and company profiling; JWT-authenticated sessions with 1-hour token expiry; self-service profile editing, password change with client-side strength meter, and GDPR account deletion. Admin dashboard lists all users in a searchable, filterable table with live status badges — activate/deactivate, toggle verified/admin, delete accounts with confirmation guard. Per-user avatar upload: client-side Canvas API resizes any image to 256×256 JPEG at 88% quality before upload, keeping payloads under 30 KB — stored as base64 TEXT in PostgreSQL, served via dedicated cache-controlled endpoint. Platform stats tile: total users, active/inactive split, admin count, verification rate, 30-day new registrations. Idempotent schema migrations run on every startup using ALTER TABLE IF NOT EXISTS — zero manual DB maintenance on deploy.
End-to-end request flow from browser to data sources — every layer secured, observable, and self-healing.
An EAN barcode in — enriched intelligence out. This is what Horus does in under 2 seconds.
Ready to see Horus analyze your actual product catalog?
Request Full Demo AccessRegister for early access or sign in to your existing account.
Register for early access to the Horus platform.
Access your Horus dashboard and intelligence modules.
20+ years building enterprise digital infrastructure. Former EMA, founder & CEO of europro.NET. Martech expert and transformation architect across logistics, distribution, and food wholesale in Europe. Horus is the product of two decades of watching businesses struggle with the gap between their data and their decisions. LinkedIn