Infrastructure and cloud usage
eDocify needs cloud infrastructure because it processes documents asynchronously, stores files securely, runs OCR/AI workloads, exports to ERP systems, and provides audit-ready evidence for customer workflows.
The product can run in controlled early environments, but the target architecture is Azure-ready so customer pilots can scale from MVP to production with security, monitoring, and operational discipline.
Cloud services used by eDocify
| Area | Cloud service category | eDocify usage |
|---|---|---|
| Web application | App hosting / containers | Tenant portal, verification workbench, admin workspace, reporting, and archive UI. |
| API backend | App hosting / containers | Document workflow API, admin API, mobile API, integration API, and archive API. |
| Background processing | Queue workers | OCR jobs, AI extraction, document preparation, ERP export, retry, and dead-letter handling. |
| Document storage | Object storage | Original PDFs, images, XML files, OCR outputs, preview assets, and archive copies. |
| Metadata storage | Database | Tenants, companies, users, roles, workflow status, field values, export jobs, audit records. |
| OCR / AI | Azure AI services | Document extraction, invoice recognition, layout analysis, confidence scoring, and future AI assistants. |
| External AI/OCR providers | Mistral OCR and compatible providers | Provider benchmarking, extraction comparison, fallback, and hybrid AI/OCR routes. |
| Local OCR / ML processing | Worker compute | Tesseract, RapidOCR, PaddleOCR-style local OCR, Ollama-hosted local LLM experiments, proprietary AI Learning engine jobs, .NET machine learning, candidate ranking, sensitive-document processing, and cost-controlled fallback routes. |
| Secrets | Key management | ERP credentials, API keys, AI provider keys, signing keys, and connector secrets. |
| Monitoring | Logs / metrics / tracing | Processing time, failure rates, queue depth, export success, SLA, and incident response. |
| Identity | Microsoft Entra ID / SSO | Enterprise sign-in, MFA, role mapping, and future SCIM provisioning. |
| Networking | Private endpoints / firewall rules | Secure customer environments and restricted data access. |
| Backup and recovery | Backup storage / snapshots | Restore document metadata, archive evidence, and customer configuration. |
Target Azure architecture
flowchart TB
U["Users and mobile app"] --> W["eDocify web app"]
W --> API["eDocify API"]
API --> DB["Metadata database"]
API --> S["Document object storage"]
API --> Q["OCR / export queues"]
Q --> Worker["Background workers"]
Worker --> AI["Azure AI / OCR providers"]
Worker --> LOCAL["Local OCR / .NET ML"]
Worker --> ERP["ERP connectors"]
API --> KV["Key Vault"]
API --> MON["Monitoring and audit logs"]
Worker --> MON
Why startup credits matter
Microsoft for Startups describes technical benefits such as startup credits for eligible Azure services, access to AI tools, and resources for building and scaling solutions. eDocify can use these credits directly for product development and pilots:
- Build pilot environments for accounting firms and enterprise finance teams.
- Run OCR/AI workloads against real document datasets.
- Compare Azure, Mistral, and local OCR routes against the same verified dataset.
- Test hybrid combinations such as cloud OCR + local OCR fallback, OCR + deterministic accounting rules, and AI extraction + human verification.
- Test local extraction combinations such as PDF/JPG invoice -> RapidOCR/PaddleOCR -> OCR text -> Ollama
qwen3:8borqwen3:30b-> JSON schema. - Run eDocify AI Learning jobs that convert verifier corrections into reusable extraction, ranking, confidence, and provider-routing signals.
- Run .NET machine learning experiments for candidate ranking and correction learning.
- Store documents and generated archive evidence securely.
- Benchmark providers on accuracy, speed, and cost.
- Test queue workers, retry policies, and dead-letter handling under realistic volume.
- Deploy monitored demo, staging, and pilot environments.
- Prepare enterprise security features such as SSO, Key Vault, logging, and data retention.
Reference: Microsoft for Startups.
Application alignment
The credits would be used to build and validate Bitlogika's own B2B AI SaaS product, not to fund custom development services for a single client. The intended use is product infrastructure, AI benchmarking, secure pilot environments, and repeatable deployment patterns for eDocify.
| Microsoft startup signal | eDocify evidence |
|---|---|
| Software product | eDocify is a proprietary document operations platform with web, API, mobile, OCR/AI, archive, and workflow modules. |
| AI solution | AI is used for OCR routing, document understanding, field extraction, JSON structuring, confidence, correction learning, and provider benchmarking. |
| Build and test on cloud | Azure can host pilot, staging, monitoring, storage, queues, AI workloads, and secure secrets. |
| Pilot/customer readiness | eDocify has defined pilot scenarios for accounting firms, enterprise finance teams, verification teams, and archive workflows. |
| Scalable product path | The roadmap moves from MVP and pilots to enterprise security, monitoring, billing, and partner readiness. |
Credit usage work packages
| Work package | Azure / infrastructure usage | Expected output |
|---|---|---|
| Pilot environments | App hosting, storage, database, queue workers, monitoring | Secure demo, staging, and pilot environments for real customer validation. |
| OCR/AI benchmarking | Azure AI, Mistral route comparison, local OCR worker compute, result storage | Field-level benchmark across Azure, Mistral, RapidOCR, PaddleOCR, Tesseract, hybrid routes, and local LLM experiments. |
| Local + cloud hybrid extraction | Worker compute, storage, queue orchestration, logs | Tested routes such as cloud OCR + local fallback, OCR + rules, and RapidOCR/PaddleOCR -> Ollama Qwen3 -> JSON schema. |
| AI Learning engine | Worker compute, metadata database, secure dataset storage | Verifier corrections converted into extraction, ranking, confidence, and provider-routing signals. |
| Enterprise security | Key Vault, Entra ID / SSO preparation, logs, retention storage | Security-ready pilot architecture with managed secrets, access control, audit logs, and retention policies. |
| Reliability testing | Queues, dead-letter handling, monitoring, retry policies | Measured OCR/export queue behavior under realistic document volume. |
| Customer evidence | Storage, reporting, dashboards, archive evidence | Pilot report with accuracy, export success, processing time, cost per document, and audit evidence. |
Pilot environment plan
| Environment | Purpose | Data policy |
|---|---|---|
| Local development | Engineering and fast iteration. | Synthetic or developer-approved samples only. |
| Demo sandbox | Public demos and role-based product tours. | Synthetic/demo data only. |
| Pilot staging | Customer pilot setup, connector tests, OCR benchmarking. | Customer-approved pilot data with access controls. |
| Production pilot | Limited live usage for selected customers. | Real customer data with tenant isolation, audit logs, retention policy, and monitored operations. |
Operational metrics
The infrastructure will be measured by:
- document processing time;
- OCR job queue depth;
- failed OCR jobs;
- export success rate;
- retry and dead-letter counts;
- storage growth;
- average verification time;
- approval SLA;
- archive retrieval time;
- API latency;
- per-document AI cost.
Funding-ready message
eDocify needs cloud credits for real product workloads: AI document extraction, secure file storage, asynchronous processing, monitored pilot environments, and enterprise integrations. This is infrastructure for a proprietary AI SaaS product, not hosting for a consulting project.