EndorsedendorsedCompleted

Artificial Intelligence Architecture Engineering for a Digital Storytelling Platform

Services Covered:

API Development • Custom Software Development • AI and Machine Learning
Project hero — replace with case study imagery
Industry
Industry

Arts, Entertainment & Recreation

Duration
Duration

5 months

Budget
Budget

10K - 49K

Start Date
Start Date

1 March 2026

Team Size
Team Size

1-5

Engagement Model
Engagement Model

Onshore

Project Capability Score (PCS)

gauge meter
76/100
Good

Project Capability Score (PCS) estimates how capable 4soft is of successfully delivering a project like this, based on its past experience, track record, and ability to handle similar work. Learn more about PCS


Client

Appventis

mapRadom, Poland

Project Summary

4soft designed and implemented the core AI architecture for a children's storytelling platform, building a CMS and a multi-layered Guardrails moderation pipeline to ensure child-safe content. They developed a vector embedding and retrieval system using pgvector and HNSW indexing to enable story continuity and personalized recommendations, and instituted a Promptfoo-based evaluation framework to secure and optimize LLM usage. Their work improved platform security, performance, latency, and cost-efficiency while maintaining timely agile delivery.

Key Challenges

  • Ensure Uncompromising Child Safety by building a watertight content moderation/guardrails pipeline to prevent harmful AI outputs.
  • Deliver a Cohesive and Personalized User Experience by creating an embedding and retrieval system that maintains story continuity and enables personalized recommendations.
  • Optimize AI Security and Cost-Efficiency by protecting against prompt-injection attacks and establishing an automated evaluation framework to compare model cost and quality.

Project Deliverables

  • Custom Guardrails Pipeline: A fully functional, multi-layered moderation node combining LangChain, API integrations, and a fine-tuned classifier to ensure 100% child-safe content generation.
  • Vector Embedding & Retrieval Architecture: A scalable pgvector-based database system featuring HNSW indexing and LangChain integration to power story continuity and personalized user recommendations.
  • Prompt Evaluation Framework: An automated testing environment utilizing Promptfoo to ensure prompt-injection security and enable continuous model performance and cost analysis.

Project Solution

4soft designed and implemented the end-to-end core AI architecture for a children's storytelling platform with a focus on child safety, scalability, and cost-efficiency. They engineered a Guardrails moderation pipeline integrating LangChain, external moderation APIs, and a fine-tuned classifier to filter AI-generated content. To support story continuity and personalization, they generated summaries via a lighter LLM, converted them into vector embeddings, and stored them using pgvector with HNSW indexing exposed through LangChain’s PGVector vectorstore. They also implemented a Promptfoo-based evaluation framework to automate prompt-injection testing and compare cost-to-quality tradeoffs between LLMs before production rollout.

Project Outcome

  • 100% Child-Safe Content Delivery: Achieved an exceptional block rate preventing harmful outputs from reaching users.
  • Optimized LLM Operational Costs: Identified and deployed lighter, cost-effective models for specific tasks, reducing API and token costs.
  • Enhanced Prompt Security: Automated testing successfully blocked simulated prompt-injection attacks and increased confidence before production changes.
  • Improved System Latency and User Engagement: HNSW indexing enabled low-latency similarity searches, contributing to longer session durations and higher retention.

Platforms

  • AI/ML PlatformAI/ML Platform
  • CMSCMS

Tech Stack

  • PostgreSQLPostgreSQL

Client Endorsement

Overall Review Rating

5star5 out of 5 stars

Timeliness

Cost Rating

Willing to Refer

Quality of Deliverables

4soft improved our platform's security, performance, and cost efficiency. Their agile project management supported timely delivery and prompt responses to our needs. The team's technical expertise and proactive problem solving played an important role in the project's success.

Dawid Wenderski

Co-Founder

Note: This endorsement is based on publicly available client feedback from external review sources.

More Projects by 4soft

Endorsed
Completed
Software Development

Web3 Architecture & RWA Tokenization

Preview of the Web3 Architecture & RWA Tokenization
Budget:
Confidential
Duration:
6 months
Team Size:
1-5

Services Covered:

+2 more