AIAH GenerativeAI Engine

Overview

AIAH is a digital parenting application used by thousands of mothers to access real, verified, and reliable advice for their parenting needs. The platform helps parents navigate pregnancy, baby care, early childhood development, feeding, sleep, health, and everyday parenting routines through trusted, expert-backed guidance delivered at scale.

The AIAH GenerativeAI Engine is the intelligent core that powers these AI-assisted parenting experiences. When the business needed to rebuild this engine, the objective was not simply to add conversational AI — it was to deliver grounded, specialist-verified parenting guidance through a secure, governable, and cost-aware architecture that could evolve alongside growing user demand and continuously changing parenting content.

Parenting expert and occupational therapist Meg Faure, founder of the organisation behind AIAH, took close interest in curating each piece of documentation that entered the AI engine. This placed verified specialist knowledge at the centre of the platform's AI design, ensuring that mothers received advice grounded in trusted content rather than generic model responses.

The Challenge

As AIAH continued to scale, the platform needed a rebuilt AI engine capable of delivering reliable parenting guidance while meeting strict requirements around content trust, operational governance, performance, and cost control.

The existing approach could not support the level of control, transparency, and flexibility the business required. Several critical challenges shaped the engagement:

Delivering AI advice exclusively from verified specialist documentation, not unverified or generic sources
Ringfencing data and inference within a trusted AI provider to prevent client information or user questions from reaching public internet-hosted models
Supporting flexible AI configuration across multiple models and prompt versions for ongoing performance evaluation
Enabling user-friendly ingestion and management of verified knowledge into the RAG engine through an admin portal
Organising the knowledge corpus so documents could be added, removed, and re-indexed in real time without development effort
Optimising AI cost and response latency, with full visibility into every AI call from ingestion through to response
Building evaluation workflows to benchmark prompts, models, and configurations against hundreds of expected user questions
Establishing a robust DevOps pipeline with Infrastructure as Code across multiple environments

AIAH therefore required a secure, governable, and performance-optimised GenerativeAI engine that could deliver trusted parenting guidance at scale while giving administrators full control over knowledge, AI configuration, cost, and quality.

The Solution

To meet AIAH's requirements, Tati Software designed and rebuilt the GenerativeAI Engine as a governed AI platform built around verified knowledge retrieval, flexible model orchestration, administrative control, cost awareness, and continuous evaluation.

The solution introduced a layered architecture that separated the engine into dedicated capabilities for:

Trusted AI inference through AWS Bedrock
Verified knowledge ingestion and retrieval
Flexible AI model and prompt configuration
Administrative governance and real-time content management
Cost tracking and performance optimisation
Evaluation and benchmarking workflows
DevOps and multi-environment infrastructure delivery

This architecture gave AIAH a production-ready GenerativeAI engine capable of delivering specialist-verified parenting guidance while maintaining strong operational control, transparency, and flexibility for future AI innovation.

Modern Identity & Security Architecture

One of the most important client requirements was ensuring that AI advice could never originate from unverified sources or leak sensitive user context into public inference environments.

Tati Software ringfenced the AI engine within AWS Bedrock as the trusted AI model supplier. This ensured that inference remained within a governed AWS environment rather than routing prompts to generic publicly hosted AI engines where client information or user questions could be exposed to the internet.

The platform was designed so that parenting advice could only be generated from verified specialist documentation curated and approved for use in the engine. Meg Faure and specialist reviewers maintained close oversight of the knowledge corpus, ensuring that every document entering the AI engine met the organisation's standards for accuracy, relevance, and trust.

This trusted AI architecture gave AIAH confidence that:

Parenting advice was grounded in verified specialist content
User questions and client data remained within a controlled inference boundary
The organisation could not inadvertently deliver guidance from unverified or generic AI sources
Knowledge governance and AI security were aligned from the outset

By combining verified content curation with ringfenced AWS Bedrock inference, Tati Software established a trusted foundation for AI-assisted parenting guidance at scale.

Intelligent AI & Retrieval Architecture

To support intelligent and adaptable parenting assistance, Tati Software rebuilt the GenerativeAI Engine around a structured retrieval-augmented generation (RAG) pipeline and flexible AI orchestration layer.

The engine was designed with flexibility for AI configuration at its core. Administrators could manage multiple AI models and prompt versions, enabling the business to evaluate how different configurations performed against real user questions over time. Rather than locking the platform to a single model or prompt, the architecture supported controlled experimentation and performance comparison across configurations.

Instead of sending prompts directly to language models, the platform implemented a structured retrieval pipeline that:

Determined user intent and parenting context
Identified relevant verified knowledge collections
Retrieved contextual information from vector-enabled databases
Ranked and filtered retrieved content for relevance and quality
Constructed grounded prompts from verified specialist documentation
Applied safety and response controls before delivery

The platform also introduced streaming responses to reduce perceived wait time for users, delivering parenting guidance incrementally as the model generated output. Combined with parallel processing, caching, and background task optimisation across the critical path, this significantly improved the user experience for mothers seeking immediate, reliable advice.

This architecture allowed AIAH to deliver accurate, explainable, and personalised AI-generated parenting guidance grounded exclusively in verified specialist knowledge.

Knowledge Management & AI Governance

A major focus of the solution was establishing governed AI knowledge workflows rather than unmanaged vector storage.

Tati Software designed and implemented a structured knowledge ingestion and management framework accessible through the admin portal, allowing non-technical administrators to manage the AI engine's knowledge base in real time. Any authorised user could see which documents were currently being used by the engine, add new verified content, remove outdated material, and trigger vector index updates without requiring further development effort.

The knowledge corpus was organised to support:

Upload and categorisation of verified specialist documentation
User-friendly document ingestion through the admin portal
Real-time addition and removal of knowledge assets
Automatic vector index updates when the corpus changed
Visibility into which documents were active in retrieval
Embedding pipeline management and re-indexing workflows
Version control and archival of knowledge assets

The platform also introduced formal AI governance tooling that enabled administrators to manage:

Prompt versions and agent configurations
Model selection and retrieval parameters
Evaluation datasets and benchmark testing workflows
Fallback logic and configuration permissions
Continued monitoring of AI performance against baseline benchmarks

This governance layer ensured AI quality could be continuously monitored, evaluated, and improved — both before and after production deployment — through tooling available directly in the admin portal.

Technical Architecture

The GenerativeAI Engine was built with cost awareness and user experience optimisation as first-class design principles from day one.

Tati Software measured every stage of the response critical path — context retrieval, authentication, model inference, and response delivery — and introduced efficiency layers throughout the architecture. These included caching, parallel task execution, background processing, and minimisation of the critical path to response. Through this work, typical user response times were reduced from over 10 seconds to under 3 seconds in general operation, with streaming responses further decreasing perceived wait periods for mothers using the application.

Cost visibility was embedded at every layer of the platform. Each and every AI call was costed, logged, and measured, giving AIAH 100% cost visibility across data ingestion, RAG retrieval, and AI response generation. This granular tracking extended to the micro-cent level, enabling the business to compare the cost of different model and prompt configurations as part of ongoing evaluation workflows.

The solution was deployed using a secure AWS cloud-native environment with a robust DevOps pipeline defining Infrastructure as Code across multiple environments. The architecture included:

Amazon ECS for containerised AI and application services
PostgreSQL with pgvector for vector-enabled retrieval workloads
Redis caching for low-latency session and retrieval data
Application Load Balancers and secure VPC networking
AWS Bedrock for governed AI model inference
Amazon S3 for document storage and knowledge assets
IAM and Secrets Manager for access control and credential management
Infrastructure as Code for repeatable multi-environment deployment

Tati Software also designed the platform with operational resilience in mind, including support for autoscaling, distributed tracing, structured monitoring, and controlled release practices across development, staging, and production environments.

Business Impact

The AIAH GenerativeAI Engine rebuild established a trusted, governable, and performance-optimised foundation for AI-assisted parenting experiences at scale.

Through this solution, Tati Software enabled AIAH to:

Deliver verified, specialist-backed parenting guidance through a ringfenced AWS Bedrock AI engine
Give administrators full control over knowledge content through a user-friendly admin portal
Support flexible AI model and prompt configuration for ongoing performance evaluation
Reduce response times from over 10 seconds to under 3 seconds with streaming delivery
Achieve 100% cost visibility across all AI operations down to the micro-cent
Evaluate prompts, models, and configurations against hundreds of expected user questions
Operate a robust DevOps pipeline with Infrastructure as Code across multiple environments
Maintain continued monitoring of AI performance against baseline benchmarks over time

By combining trusted AI inference, governed knowledge management, cost-aware architecture, and rigorous evaluation workflows, Tati Software helped AIAH build a GenerativeAI engine capable of delivering reliable parenting guidance to thousands of mothers while maintaining the operational control and transparency the business required.

Metrics for Success

KPI 1: AI Response Latency

Baseline:
Before the engine rebuild, users often experienced AI response times exceeding 10 seconds, creating friction for mothers seeking immediate parenting guidance and reducing confidence in the platform's responsiveness.

Target:
Reduce end-to-end AI response latency to under 3 seconds for typical user interactions while introducing streaming responses to further decrease perceived wait time.

Measured Result:
Typical user response times were reduced from over 10 seconds to under 3 seconds in general operation. Streaming responses were introduced to deliver incremental output to users, further decreasing perceived wait periods during AI-generated parenting guidance.

Measurement Method:
Comparison of pre- and post-rebuild response time measurements across context retrieval, authentication, model inference, and response delivery stages, including critical path analysis and user-facing latency benchmarks.

Business Impact:
Mothers using the AIAH application received faster, more responsive parenting guidance, improving user experience and confidence in the platform's ability to deliver timely, reliable advice.

KPI 2: AI Cost Visibility and Configuration Evaluation

Baseline:
Before implementation, the platform lacked granular cost tracking across AI operations, making it difficult to compare the cost and performance of different model and prompt configurations or to evaluate AI spending with precision.

Target:
Establish 100% cost visibility across all AI operations — from data ingestion through RAG retrieval to model inference — enabling cost comparison of configurations at the micro-cent level and supporting ongoing evaluation workflows.

Measured Result:
The platform achieved 100% cost visibility across data ingestion, RAG calls, and AI responses. Every AI call was costed, logged, and measured, enabling administrators to compare the cost of different model and prompt configurations to the micro-cent as part of evaluation and benchmarking workflows.

Measurement Method:
Review of AI cost logging, per-call measurement records, configuration comparison reports, and evaluation workflow outputs demonstrating granular cost tracking across all AI pipeline stages.

Business Impact:
AIAH could make informed decisions about AI model and prompt configurations based on both performance and cost, supporting sustainable scaling of AI-assisted parenting guidance while maintaining operational transparency.

Outcomes

The AIAH GenerativeAI Engine rebuild created a trusted, governable, and performance-optimised AI platform for verified parenting guidance. The architecture ringfenced inference within AWS Bedrock, gave administrators real-time control over verified knowledge content, reduced response latency dramatically, and established full cost visibility and evaluation workflows for continued AI quality monitoring.

The client was extremely happy with Tati Software's delivery, particularly the combination of trusted AI architecture, administrative governance, performance optimisation, and robust DevOps practices that gave the business confidence to scale AI-assisted parenting experiences to thousands of users.

Lessons Learned

Key lessons included the importance of ringfencing AI inference within trusted providers when client data and content trust are paramount, and the value of building cost awareness and evaluation workflows into the platform from day one rather than retrofitting them later.

The engagement also reinforced the importance of user-friendly knowledge management through administrative portals, flexible model and prompt configuration for ongoing evaluation, and Infrastructure as Code DevOps pipelines for reliable multi-environment delivery.

Tati Software will continue applying these lessons to future GenerativeAI engagements by prioritising trusted inference boundaries, granular cost visibility, governed knowledge workflows, performance-critical path optimisation, and evaluation tooling that supports continued monitoring against baseline performance over time.