Sentiant-RO 🤖 - Romanian AI Agent on Masumi Network

[Proposal setup] Proposal title

Please provide your proposal title

Sentiant-RO 🤖 - Romanian AI Agent on Masumi Network

[Proposal Summary] Budget Information

Enter the amount of funding you are requesting in ADA

52000

[Proposal Summary] Time

Please specify how many months you expect your project to last

8

[Proposal Summary] Translation Information

Please indicate if your proposal has been auto-translated

No

Original Language

en

[Proposal Summary] Problem Statement

What is the problem you want to solve?

Romanian e-commerce lacks AI-powered sentiment analysis tools, limiting businesses' ability to understand customer feedback and make data-driven decisions in their native language.

[Proposal Summary] Supporting Documentation

Supporting links

• https://rocsi.eu/
,
• https://docs.google.com/document/d/1G6r39PN_0icZW0bhqck62shZsCxQXh6oaRB3GIW0Phc/

[Proposal Summary] Project Dependencies

Does your project have any dependencies on other organizations, technical or otherwise?

Yes

Describe any dependencies or write 'No dependencies'

1. Masumi Network Platform: Core dependency for AI agent deployment, payment processing, and Cardano integration. 2. Google Cloud Vertex AI: Essential for model training pipeline and MLOps infrastructure. 3. LLM API Services: GPT-4/Claude APIs required for automated labeling of 100k Romanian reviews. 4. Romanian Data Sources: Access to Romanian e-commerce review data for model training. 5. Cloud Infrastructure: Google Cloud Run/Railway for service hosting with auto-scaling capabilities.

[Proposal Summary] Project Open Source

Will your project's outputs be fully open source?

Yes

License and Additional Information

All project code will be open source under MIT: data collection/cleaning & labeling pipelines, Vertex AI training/eval, FastAPI Masumi service, Docker/Helm/Terraform configs, and CI. Docs under CC BY 4.0. We won’t redistribute review text; instead we’ll publish dataset schema, cleaning/labeling scripts, metrics, guidelines, a small lawful sample, and (where lawful) trained model/adapters.

[Theme Selection] Theme

Please choose the most relevant theme and tag related to the outcomes of your proposal.

E-commerce

[Campaign Category] Category Questions

Describe what makes your idea innovative compared to what has been previously funded (whether by you or others).

Sentiant-RO is the first Romanian-language AI sentiment analysis service on blockchain infrastructure. Unlike existing tools that primarily support English, we combine:

1. Specialized training on 100k Romanian reviews
2. Masumi Network integration for ADA payments
3. Focus on underserved Romanian e-commerce

No previous Catalyst projects addressed Romanian language AI service monetization through Cardano. This validates a new model where specialized language AI services can operate on blockchain infrastructure.

Describe what your prototype or MVP will demonstrate, and where it can be accessed.

Our MVP demonstrates a functional Romanian sentiment analysis AI agent on Masumi Network.

The prototype showcases:

1. Real-time Romanian text analysis with >85% accuracy
2. ADA payment processing through Masumi infrastructure
3. Web dashboard with usage analytics
4. API documentation for e-commerce integration

Accessible via dedicated Masumi Network agent integration URL, validating feasibility and commercial viability of Cardano AI services.

Describe realistic measures of success, ideally with on-chain metrics.

Success measured through on-chain and technical metrics:

1. 10 successful ADA payments for sentiment analysis within 30 days of launch
2. 100 ADA in total service revenue demonstrating market demand
3. 85% sentiment analysis accuracy on Romanian text
4. 99% uptime with <500ms API response times
5. Service listed via Masumi Network with mostly positive user ratings.

These metrics validate technical feasibility of Romanian AI on Cardano and commercial viability of blockchain-native AI services, establishing foundation for broader language-specific AI adoption.

[Your Project and Solution] Solution

Please describe your proposed solution and how it addresses the problem

Sentiant-RO addresses the lack of Romanian-language AI services in the Cardano ecosystem by creating the first blockchain-native sentiment analysis agent.

Our solution tackles three problems:

1. Language Gap – No specialized NLP tools exist for Romanian sentiment analysis, leaving businesses with inaccurate multilingual services.
2. AI Monetization – AI services are typically locked to fiat systems, limiting blockchain adoption.
3. Market Need – Romanian e-commerce (worth €6.2B and growing) lacks automated tools for customer feedback analysis.

Technical Approach

• Data Layer: Collect and curate 100,000 Romanian e-commerce product reviews (PII-free), clean and categorize them for training.
• AI/ML Layer: Use Google Cloud Vertex AI to fine-tune Romanian BERT/XLM-R models, with GPT-4/Claude pre-labeling and native speaker validation to ensure cultural and linguistic accuracy.
• Service Layer: Deploy a FastAPI microservice with Masumi-compliant endpoints, auto-scaling, and real-time sentiment inference (<500ms).
• Blockchain Integration: Integrate with the Masumi Network for ADA-based payments, identity, and decision logging—enabling transparent and decentralized access to AI.

Problem Resolution

• Enables culturally accurate Romanian sentiment analysis (e.g., understanding idiomatic expressions like “merge ca unsă”).
• Demonstrates AI monetization natively through Cardano ADA.
• Provides Romanian businesses with affordable, automated insights for decision-making.

Innovation & Value

• Technical: First integration of Romanian AI services with blockchain-native payments.
• Economic: Lowers barriers for local businesses to adopt AI-driven analytics.
• Ecosystem: Expands Masumi/ Cardano adoption into Romanian markets and provides a replicable model for other underserved languages.

Implementation Phases

1. Data Foundation (M1-2): Ethical collection & validation of 100k Romanian reviews.
2. AI Development (M3-4): Automated labeling + human validation, Vertex AI model training.
3. Integration (M5-6): FastAPI service, Masumi SDK endpoints, ADA payments.
4. Deployment (M7-8): Marketplace launch, documentation, user feedback & validation.

[Your Project and Solution] Impact

Please define the positive impact your project will have on the wider Cardano community

Sentiant-RO creates lasting positive impact for the Cardano community through seven key channels:

• Community Inclusion: Adds Romanian language support, opening Cardano to 19M speakers and a growing tech sector. This signals commitment to accessibility and inspires other language communities.
• Developer Ecosystem Growth: Provides a fully open-source AI service template (MIT license) for language-specific agents, encouraging replication in other markets.
• Technical Innovation: Demonstrates real-world integration of Vertex AI with the Masumi Network, showcasing Cardano’s ability to host advanced AI services.
• Economic Activity: Every sentiment query drives ADA usage and sustainable revenue, while Romanian businesses become active Cardano stakeholders.
• Education & Knowledge Sharing: Documentation, tutorials, and a working open-source repo build community skills in AI–blockchain integration.
• Competitive Advantage: Positions Cardano as the first blockchain with Romanian AI services, setting a replicable model for Eastern Europe and beyond.
• Open Source Multiplier: MIT-licensed outputs enable community-driven improvements and derivative projects, amplifying the value of Catalyst funding.

Overall, Sentiant-RO strengthens Cardano as a platform for real-world AI applications while fostering inclusion, adoption, and sustainable growth.

[Your Project and Solution] Capabilities & Feasibility

What is your capability to deliver your project with high levels of trust and accountability? How do you intend to validate if your approach is feasible?

CAPABILITY TO DELIVER WITH TRUST & ACCOUNTABILITY

Proven Track Record:

• Certified Cardano Blockchain Associate with deep ecosystem knowledge.
• Funded Catalyst participant (Fund 13) with transparent progress reporting.
• 20+ years’ experience across software development, AI, blockchain, and cloud, from VB6/.NET to AI and Web3.
• Native Romanian speaker with cultural/linguistic expertise critical for sentiment model accuracy.

Relevant Experience:

• Developer of NMKR WordPress plugin, demonstrating blockchain API integration with user platforms.
• Advisory Board Member at VeterinaryDAO, contributing to decentralized governance and AI adoption.
• Former advisor/manager in SingularityNET ecosystem projects (Deep Funding, Rejuve.AI, Cogito, etc.), evidencing AI service marketing and community engagement.

Accountability Mechanisms:

• Open-source codebase from inception with public repository.
• Transparent milestones with specific deliverables, acceptance criteria, and progress reports.
• Established reputation and ongoing engagement in the Cardano community.

VALIDATION OF FEASIBILITY

Technical Validation:

• Model training on Google Vertex AI, a proven platform for multilingual NLP.
• Masumi Network integration via documented SDK, already supporting live AI agent deployments.
• PII-free Romanian product review dataset confirmed feasible (100k reviews from multiple platforms).

Market Validation:

• Romanian e-commerce: €6.2B, 15% annual growth, 8.4M users—clear demand for native AI tools.
• No Romanian-specific sentiment solutions currently available, confirming market gap.

Risk Mitigation:

• Redundant data sources across 3+ platforms.
• Alternative training environments (AWS/Azure) if GCP issues arise.
• Progressive milestone delivery allowing course correction.
• Continuous validation by Romanian native speakers.

Progressive Milestone Validation:

• Dataset quality & compliance validation.
• LLM pre-labeling accuracy checks.
• Model benchmarking (>87% accuracy target).
• Masumi integration with live ADA transactions.
• Market validation through real user adoption.

Financial Feasibility:

• Transparent budget with 20% contingency buffer.
• Cost-effective solo developer approach with support network (community validators, cloud support, Masumi ecosystem).
• Long-term sustainability through ADA transaction revenues.

With demonstrated expertise, transparent accountability practices, and phased feasibility validation, the project is positioned for reliable delivery and high trust within the Cardano community.

[Milestones] Project Milestones

Milestone Title

Romanian Product Review Dataset Creation

Milestone Outputs

• Cloud-hosted structured dataset with review storage and categorization
• Data cleaning & validation pipeline with automated PII removal
• 100,000 Romanian product reviews dataset (PII-free and categorized)
• Data quality metrics report showing distribution across product categories
• Public GitHub repository with documented data collection scripts and methodologies

Acceptance Criteria

• Dataset is deployed on a cloud-hosted service with defined schema (review text, product category, sentiment label, confidence level) and allows query access
• Pipeline executes end-to-end on raw data, with automated checks ensuring zero PII and logs for each cleaning step
• Final dataset contains ≥100,000 unique reviews, with PII-free content and category field populated for ≥95% of entries
• Report demonstrates balanced coverage across ≥8 product categories, <5% duplicates, and >90% usable content
• Public GitHub repository is accessible, contains all scripts, and includes a README with instructions to reproduce at least a 1,000-sample subset

Evidence of Completion

• Cloud Storage dataset export with anonymized sample reviews and schema snapshot showing fields (review text, product category, sentiment label, confidence level)
• Vertex AI data processing job logs stored in Cloud Storage, including PII-scan validation report showing 0 flagged entries
• Row count report stored in Cloud Storage and random sample inspection confirming dataset size and completeness
• Cloud Storage analytics report including category distribution charts, deduplication statistics, and usability metrics generated via Vertex AI jobs
• Public GitHub repository URL with commit hash showing data collection scripts, cleaning pipeline, and README documentation

Delivery Month

2

Cost

10500

Progress

20 %

Milestone Title

Automated Sentiment Labeling System

Milestone Outputs

• LLM-powered sentiment labeling pipeline (GPT-4/Claude integration)
• 100,000 Romanian reviews enriched with sentiment labels and confidence levels (schema applied: review text, product category, sentiment label, confidence level)
• Quality validation framework with native Romanian speaker verification
• Batch processing optimization for cost-efficient API usage, with cost analysis and monitoring reports
• Export format compatible with Vertex AI training requirements
• Public GitHub repository update including labeling pipeline scripts, configuration documentation, and a reproducible sample subset of labeled data

Acceptance Criteria

• Functional labeling pipeline is deployed and can process raw reviews into labeled outputs with confidence levels
• Final dataset contains ≥100,000 reviews labeled with both sentiment and confidence level in the agreed schema
• Validation by native Romanian speakers confirms ≥80% labeling accuracy on 3,000 manually labeled samples
• Inter-annotator agreement achieves ≥0.75 Cohen’s kappa across validation subset
• Batch processing achieves average cost ≤0.12 ADA per review and produces usage monitoring reports
• Public GitHub repository is updated with scripts/notebooks for the labeling pipeline and includes a reproducible example subset of labeled data (e.g., 1,000 rows)

Evidence of Completion

• Vertex AI labeling pipeline job logs stored in Cloud Storage showing successful labeling runs and configuration documentation
• Cloud Storage export of labeled dataset in Vertex AI-compatible format with schema (review text, product category, sentiment label, confidence level)
• Validation report stored in Cloud Storage with accuracy metrics and results of native speaker review (3,000 manually labeled samples)
• Inter-annotator agreement report (Cohen’s kappa) stored in Cloud Storage and generated from validation subset
• Cloud Storage cost optimization report from Vertex AI batch processing jobs, with ADA cost metrics and usage monitoring details
• Public GitHub repository URL with commit hash showing labeling scripts, documentation, and a reproducible subset of labeled data

Delivery Month

4

Cost

14000

Progress

40 %

Milestone Title

Romanian Sentiment Analysis Model

Milestone Outputs

• Production-ready Romanian sentiment analysis model with >83% accuracy
• Vertex AI MLOps pipeline with automated training, versioning, and deployment workflows
• Model performance evaluation across multiple Romanian product categories
• Compressed and optimized model for efficient production deployment
• Benchmarking report against baseline models
• Cross-domain validation reports showing consistent performance metrics
• Public GitHub repository update including inference demo scripts, sample input/output data, and usage documentation

Acceptance Criteria

• Model achieves >83% accuracy on Romanian sentiment classification tasks using dataset schema (review text, product category, sentiment label, confidence level)
• Automated Vertex AI pipeline successfully trains on the Romanian dataset end-to-end, with reproducible versioning and deployment automation
• Performance consistency achieved across ≥5 product categories with variance under 10%
• Final model size optimized to ≤150MB for efficient deployment
• Model performance exceeds selected baseline(s) (e.g., multilingual or translation-based sentiment model) by ≥5% accuracy improvement or achieves state-of-the-art performance for Romanian sentiment classification
• Cross-domain validation demonstrates <12% performance degradation across domains
• Public GitHub repository is updated with a runnable inference demo script, a sample subset of labeled reviews (e.g., 100 rows), and README usage instructions

Evidence of Completion

• Vertex AI training logs and Cloud Storage model artifacts showing achieved accuracy >83%, with detailed performance evaluation report (accuracy, confusion matrices, statistical analysis)
• Vertex AI pipeline documentation and logs demonstrating automated training, versioning, and deployment workflows, with model versions registered in Vertex AI Model Registry
• Vertex AI evaluation job results and stored reports in Cloud Storage, including category-specific accuracy metrics and variance calculations
• Vertex AI custom training job logs running model compression/optimization (e.g., Hugging Face Optimum / ONNX / TensorRT inside GCP container), with optimized model artifact in Cloud Storage (≤150MB) and inference benchmarks from Vertex AI Endpoint
• Benchmarking report stored in Cloud Storage comparing Romanian sentiment model against baseline(s), with metrics tables and performance graphs generated from Vertex AI evaluation jobs
• Cross-domain validation report produced from Vertex AI evaluation jobs, stored in Cloud Storage, including degradation metrics across Romanian product categories
• Public GitHub repository URL with commit hash showing inference demo script, sample input/output data (subset of reviews), and updated README documentation

Delivery Month

6

Cost

16000

Progress

70 %

Milestone Title

Kodosumi Deployment & Masumi Payment Integration

Milestone Outputs

• FastAPI-based sentiment analysis agent deployed via Kodosumi runtime with working API endpoints
• Dockerized deployment with GCP auto-scaling and Kodosumi runtime compliance validation
• Cardano ADA payment integration through the Masumi protocol SDK with identity-linked transactions
• End-to-end testing of sentiment analysis + ADA payments using Masumi protocol flows
• Developer documentation covering Kodosumi deployment setup and Masumi payment integration
• Performance monitoring and analytics dashboard showing real-time usage and health metrics via GCP

Acceptance Criteria

• Agent API is deployed and accessible via Kodosumi runtime, with endpoints validated against runtime requirements
• Service runs in Docker, supports GCP auto-scaling, and passes load-balancing validation under test load
• Successful ADA transactions are processed through the Masumi protocol, with identity authentication enabled
• End-to-end integration tests confirm sentiment analysis requests and ADA payment flows complete successfully
• Developer documentation is published and validated internally for correctness and reproducibility
• Monitoring dashboard demonstrates uptime ≥99%, latency <1.2s under load, and ability to handle ≥50 concurrent requests

Evidence of Completion

• Kodosumi runtime deployment logs stored in GCP Cloud Storage, including endpoint validation reports
• Docker image in GCP Artifact Registry with GCP deployment logs showing auto-scaling and load-balancer activity
• Masumi protocol transaction logs stored in Cloud Storage, demonstrating successful identity-linked ADA payments
• End-to-end integration test logs from GCP Cloud Logging confirming workflow completion (sentiment + payment)
• Developer documentation published in GitHub/GCP Storage, including Kodosumi deployment setup and Masumi SDK usage examples
• Performance monitoring dashboard screenshots + GCP Cloud Monitoring exports showing uptime, latency, and concurrency metrics

Delivery Month

7

Cost

9000

Progress

90 %

Milestone Title

Production Launch & Market Validation

Milestone Outputs

• Live Sentiant-RO agent listed and discoverable on the Sōkosumi marketplace, compliant with Masumi protocol standards
• FastAPI-based agent deployed via Kodosumi runtime with auto-scaling and reliable performance
• Cardano payment integration through the Masumi protocol with authenticated identities and ADA transaction support
• End-to-end testing with live ADA transactions and Masumi-compliant payment verification
• Comprehensive developer documentation and integration tutorials covering Sōkosumi listing, Kodosumi deployment, and Masumi payment flows
• Performance monitoring and analytics dashboard showing real-time usage and health metrics via GCP
• Community feedback system integrated for continuous improvement during production launch
• Project Close-out Report (PCR) detailing all milestones, outputs, and achievements in compliance with Catalyst guidelines
• Project Close-out Video (PCV) showcasing the solution’s key features, benefits, and project outcomes

Acceptance Criteria

• Service is successfully listed and searchable on the Sōkosumi marketplace with Masumi-compliant metadata
• Agent is deployed via Kodosumi runtime, achieving stable auto-scaling under test load
• ADA payments are successfully processed through the Masumi protocol with identity-linked transactions
• End-to-end integration tests confirm successful transaction flow, payment verification, and sentiment API responses
• Developer documentation is published and validated internally for correctness and ease of use
• Monitoring dashboard displays uptime ≥99%, latency <1.2s under load, and ≥50 concurrent requests supported
• Community feedback system collects and categorizes ≥20 actionable inputs from early adopters
• Project Close-out Report (PCR) is submitted to Catalyst in the required format, covering all project milestones and results
• Project Close-out Video (PCV) is published through official project and Catalyst channels, demonstrating key features and outcomes

Evidence of Completion

• Sōkosumi marketplace listing with screenshots showing discoverability and Masumi protocol compliance
• Kodosumi runtime deployment logs stored in GCP Cloud Storage, including auto-scaling test results
• Masumi protocol transaction logs stored in Cloud Storage showing authenticated ADA payments linked to agent identity
• End-to-end test reports from GCP Cloud Logging demonstrating successful workflow completion (sentiment analysis + payment)
• Public developer documentation hosted in GitHub/GCP Storage bucket with integration tutorials for Sōkosumi/Kodosumi/Masumi
• GCP Cloud Monitoring dashboard exports showing uptime, latency, and concurrency metrics
• Community feedback analysis report stored in Cloud Storage, summarizing ≥20 categorized improvement suggestions
• Final Project Close-out Report (PCR) in PDF format submitted to Catalyst and archived in Cloud Storage
• Final Project Close-out Video (PCV) published via official Catalyst/project channels, with links/screenshots as proof

Delivery Month

8

Cost

3000

Progress

100 %

[Final Pitch] Budget & Costs

Please provide a cost breakdown of the proposed work and resources

Budget Breakdown – ₳52,000 Requested

Milestone 1 – Dataset Creation (₳10,500)

• Cloud infrastructure (scraping, storage, backup)
• Data cleaning & validation pipeline (PII removal, QA)
• 100k structured Romanian product reviews dataset
• Documentation & GitHub repo with scripts

Milestone 2 – Automated Labeling (₳14,000)

• GPT-4/Claude API usage (100k reviews)
• Cost-optimized batch labeling + monitoring
• Native Romanian validation (3k samples, inter-annotator agreement)
• Export formats & pipeline documentation

Milestone 3 – Sentiment Model (₳16,000)

• Vertex AI training & hyperparameter optimization
• Romanian BERT fine-tuning, compression, optimization
• Evaluation & benchmarking across domains
• Inference-ready model with usage scripts

Milestone 4 – Deployment & Masumi Integration (₳9,000)

• Kodosumi runtime deployment with auto-scaling/load balancing
• Masumi SDK payment integration (identity-linked ADA flows)
• End-to-end testing under load (>50 concurrent requests)
• Developer documentation & monitoring dashboard

Milestone 5 – Launch & Market Validation (₳3,000)

• Public launch on Sōkosumi marketplace
• ADA payment validation & monitoring setup
• Community feedback system, demo video, docs
• Project Close-out Report & Video (Catalyst compliant)

Resource Allocation

• Technical development: 65% (₳33,800)
• Infrastructure & cloud: 20% (₳10,400)
• External services (APIs, legal): 10% (₳5,200)
• Documentation & community: 5% (₳2,600)

Risk Mitigation Buffer

Built into milestones to cover API cost variability, additional QA/validation, scaling needs, and extended training cycles.

[Final Pitch] Value for Money

How does the cost of the project represent value for the Cardano ecosystem?

The ₳52,000 investment in Sentiant-RO delivers strong value by creating the first Romanian-language sentiment AI service on Cardano, opening access to a €6.2B e-commerce market (8.4M users). The project leverages Masumi for compliance, identity, and ADA-native payments, publishes as a service in the Sokosumi marketplace, and runs on Kodosumi’s managed runtime for scale. Together, these layers showcase Cardano’s leadership in combining AI and blockchain.

Outputs & Value

• Technical: Production-ready Romanian sentiment model (>87% accuracy, <500ms inference), deployed as a Masumi-compliant API.
• Ecosystem: Open-source codebase (MIT), reusable pipelines, replicable framework for other languages.
• Market: At least 50 paid transactions at launch, proving blockchain-native AI monetization.

Compared to traditional AI development (200k+ ADA equivalent), this project achieves the same with a fraction of the cost by leveraging open models, Vertex AI efficiency, and blockchain-native infra. Beyond Romania, it establishes a replicable template for other underserved languages, multiplying ecosystem impact.

In short: Sentiant-RO turns ₳52,000 into a sustainable AI service that drives ADA adoption, generates revenue, and demonstrates the combined strengths of Masumi, Sokosumi, and Kodosumi in bringing AI agents to the Cardano ecosystem.

[Required Acknowledgements] Consent & Confirmation

Terms and Conditions:

Yes