Akiko
Genesis: from lab to reality
How Akiko was born — and why we chose to build an independent artificial intelligence.
The beginning
In February 2025, Project Nicky was launched — digitalCode’s dedicated laboratory for artificial intelligence research. The goal was clear: develop proprietary models, free from external dependencies, designed to guarantee absolute privacy and full control.
The philosophy
Zero compromises: no third-party services, no data leakage, no black boxes. Everything runs in-house or on digitalCloud, with hardware calibrated for real-world usage. Every line of code is traceable, verifiable, and fully under our control.
The evolution
Several projects emerged from the Nicky Lab: Akiko (AI chatbot), Cassandra (multimedia generation), Goliath (predictive analytics), and I.R.I.N.A. (multi-AI translation). Akiko was the first to move beyond experimentation, becoming a stable, production-ready system designed for real operational workflows.
What we built
The engineering results that make Akiko unique in the AI solutions landscape.
Fully in-house architecture
Every system layer — from the AI model to the interface — was designed and developed internally. Nothing is outsourced; every component is optimized for efficiency, security, and control.
- Custom ingestion pipelines for every source type (documents, URLs, APIs)
- Semantic retrieval system optimized for precision and speed
- Orchestration layer with built-in governance policies
Zero external dependencies
Akiko operates in complete isolation: no calls to third-party services, no data sharing. Everything runs on digitalCloud or on dedicated on-premise infrastructure.
- No vendor lock-in
- Full control over costs, performance, and security
- Guaranteed data residency: data never leaves the perimeter
RAG with verified citations
Akiko answers exclusively based on the provided content. Every response includes its source, ensuring full transparency and traceability.
Zero hallucinations
If it doesn’t know, it says so. If it answers, it cites the source.
Native multilingual capability
Understands and responds in more than 8 languages without external translation modules, adapting tone and context to the user’s language.
Experimental results
Concrete data collected during internal testing and comparisons with commercial solutions.
Response accuracy
Evaluation on a corporate knowledge base (over 500 textual documents).
Average response time
End-to-end latency measured on complex queries across different datasets and hardware configurations.
Zero hallucinations
Akiko responds only using knowledge base content and always cites the source.
Comparison with commercial solutions
What already works well
- RAG with precise, verifiable citations
- Automated and scalable ingestion from heterogeneous sources
- Native multilingual management without external modules
- Flexible deployment: on-premise or on digitalCloud
- Consistent performance even on moderate hardware
What we are refining
- Semantic ranking optimization for ambiguous queries
- Expansion of enterprise connectors
- Admin console for source management and monitoring
- Feedback loop and continuous self-improvement system
Potential application scenarios
Where Akiko delivers real value: sectors, contexts, and use cases already ready for deployment.
Public Administration
Citizen assistance and internal knowledge management for municipal and regional administrations.
Companies & Enterprises
Know-how management and automated operational support for complex organizations.
Professional Firms
Assistance with procedures, regulations, and documentation for legal, tax, and technical firms.
From the R&D Lab
Akiko is just the beginning. Every project born in our lab is a step forward toward truly independent artificial intelligence.
The R&D project exploring the foundations of private artificial intelligence.
Multi-AI translation system with 7 models and 4 filtering layers for maximum accuracy
Predictive analytics and forecasting for AI-driven financial trading
Image and music generation using proprietary in-house trained models
Explore the Lab
Discover active projects, ongoing experiments, and the philosophy behind our vision: building innovation, not just adopting it.
Go to the R&D LabTech Deep Dive
Architecture, security, performance, and integration: the engineering behind Akiko
System Architecture
The layers that make up Akiko
Ingestion & Indexing
Pipelines for connectors, document parsing, deduplication, intelligent chunking, embedding generation, and vector indexing on vector databases.
Retrieval & Ranking
Hybrid search (vector + keyword), semantic re-ranking, user-permission filtering, confidence scoring, and citation selection.
Model Layer
Custom domain-optimized LLMs, prompt engineering, context assembly, and generation with inline citations.
Orchestration & Policies
Rate limiting, prompt governance, audit logging, feedback loops, quality metrics, and an admin console for managing sources and permissions.
Security & Compliance
Privacy by design, auditing, and control
End-to-End Encryption
TLS 1.3 in transit, AES-256 at rest, secret management with Vault or equivalent solutions.
RBAC / SSO
Role-Based Access Control, LDAP/OIDC integration, and granular permissions on sources and documents.
Full Audit Trail
Immutable logs for every access, query, and change. Configurable retention policies and compliance export.
Backup & DR
Daily automated backups, tested disaster recovery, and air-gap options for critical environments.
Sizing & Performance
Hardware sized for real workloads
Small
Pilot teams / Proof of concept
- CPU + consumer-class GPU (e.g. RTX 4060 / 4070)
- 8–16 GB RAM
- 200–500 GB SSD
Medium
RecommendedBusiness unit / extended test environment
- CPU + consumer / enterprise GPU (e.g. RTX 4080 – A10)
- 32–64 GB RAM
- 1–2 TB SSD
Large
Enterprise / production
- Dedicated enterprise-class GPUs (A10 / A100 / L40)
- ≥ 64 GB RAM
- ≥ 5 TB NVMe SSD
Optimal sizing depends on concurrent users, knowledge base size, query frequency, and target latency. We assess the ideal configuration together during the initial evaluation.
Integrations
Input / output connectors
- Folders / NAS (SMB, NFS)
- Databases (PostgreSQL, MySQL, MongoDB)
- Wiki / Confluence / SharePoint
- Ticketing systems
- Email (IMAP / Exchange)
- Custom REST APIs
- JSON / CSV export
- Real-time webhooks
- Teams / Slack / Email notifications
- REST APIs for integration
Want to learn more?
Akiko is evolving, but the results already confirm the direction.
If you’re looking for an AI solution designed to guarantee independence, security, and full data control, let’s talk. We’re currently evaluating collaborations with partners interested in testing it in real operational environments.
Innovation means challenging your own ideas, not waiting for them to become standards.