Thomas BrochardDevelopment · Data · AI
Let’s talk
Back to selected work

Retrieval systems

Case study / 10

RAG Platform Architecture

Make service boundaries and delivery evidence explicit.

A proposed cloud architecture for session-aware retrieval, catalogue refresh, observability and cost planning.

Technologies & concepts

  • AWS architecture
  • FastAPI boundary
  • ECS Fargate proposal
  • OpenSearch proposal
  • Bedrock proposal
  • CloudWatch / Langfuse design

Architecture study / proposed design

Private repository

01 / The context

The problem behind the design.

A local retrieval prototype leaves session isolation, continuous indexing, operational observability and cost control unspecified. This architecture case study translates those concerns into proposed components, tradeoffs and a delivery sequence.

02 / Engineering decisions

Choices made explicit.

Separate requests from ingestion

The proposal separates API coordination, session context, filtered vector retrieval and model calls from scheduled snapshot refresh and indexing.

Document operational questions

Ownership checks, retention, retries, refresh reconciliation, provider boundaries and trace privacy are explicit design considerations. Location and date metadata act as retrieval constraints.

State acceptance and cost assumptions

A prioritized backlog defines evidence needed for delivery acceptance. Workload equations and symbolic unit rates make cost drivers inspectable without claiming delivered service levels.

03 / System view

Follow the flow.

Proposed RAG service design: a validated request uses session context and filtered retrieval, crosses a model boundary and returns a cited answer with operational signals.

Illustrative architecture from the project presentation. It describes the selected design; it does not imply a deployed service or a measured operational result.

View the full-size diagram
  1. Validated request

  2. Session context

  3. Filtered retrieval

  4. Model boundary

  5. Cited answer

  6. Operational signals

04 / Selected artifacts

What the repository contains.

  • docs/architecture.md
  • docs/decisions.md
  • docs/backlog.md
  • docs/cost-model.md

05 / Scope & limitations

How to read this project.

Architecture case study only. Service names identify proposed responsibilities; no application, infrastructure-as-code package, production configuration, endpoint or performance evidence is included. Targets and example workloads are illustrative planning inputs. Cloud compatibility, capacity, quotas and actual prices require separate verification before implementation.

Explore the source with access.

The GitHub repository is private. Its link works for authorized collaborators; other visitors can contact me to discuss the project.

Next case study / 01

BottleNeck Data Pipeline

Data engineering & automation

Explore the collection

The next conversation

Let’s talk about your project.

↗
A need, an idea, a technical question.
LinkedIn ↗GitHub ↗