🚧 UNDER CONSTRUCTION — This workshop is being actively developed. Code, content, and infrastructure may change without notice.
⚠️ Educational Workshop: Demonstration code. Not intended for production deployment without proper security hardening.
Blaize Bazaar — an e-commerce platform that starts with broken keyword search and no AI. You'll progressively add semantic search, agent tools, multi-agent orchestration, and production infrastructure by editing the real application code.
| Format | Duration | Challenges | What Participants Build |
|---|---|---|---|
| Workshop | 2 hours | 9 challenges (all edit) | Full stack: search → agents → production |
| Builder's Session | 1 hour | 2 edit + 7 test/read | Search + tools hands-on, rest pre-completed |
| Module | Name | Challenges | Outcome |
|---|---|---|---|
| 1 | Smart Search | C1: _vector_search() |
"Your database understands what customers mean." |
| 2 | Agentic AI | C2: @tool, C3: agent, C4: orchestrator |
"A multi-agent team handles customer queries." |
| 3 | Production Patterns | C5: runtime, C6: memory, C7: gateway, C8: observability, C9: identity | "Your agent system runs on managed infrastructure." |
Services auto-start on the workshop instance. If running locally:
# Terminal 1: Backend (auto-reloads on .py changes)
cd blaize-bazaar/backend
uvicorn app:app --reload --host 0.0.0.0 --port 8000
# Terminal 2: Frontend (HMR on .ts/.tsx changes)
cd blaize-bazaar/frontend
npm install && npm run devblaize-bazaar/ # The application
├── backend/
│ ├── services/
│ │ ├── hybrid_search.py ← Challenge 1: _vector_search()
│ │ ├── agent_tools.py ← Challenge 2: get_trending_products()
│ │ ├── agentcore_memory.py ← Challenge 6: AgentCore STM
│ │ ├── agentcore_gateway.py ← Challenge 7: MCP Gateway
│ │ ├── otel_trace_extractor.py ← Challenge 8: Observability
│ │ └── (12 more pre-built services)
│ ├── agents/
│ │ ├── recommendation_agent.py ← Challenge 3: specialist agent
│ │ ├── orchestrator.py ← Challenge 4: multi-agent orchestrator
│ │ ├── search_agent.py (pre-built)
│ │ ├── inventory_agent.py (pre-built)
│ │ ├── pricing_agent.py (pre-built)
│ │ └── customer_support_agent.py (pre-built)
│ ├── agentcore_runtime.py ← Challenge 5: AgentCore Runtime
│ └── app.py (pre-built FastAPI server)
├── frontend/
│ └── src/utils/agentIdentity.ts ← Challenge 9: agent identity UI
│
solutions/ # Drop-in replacements (cp and restart)
├── module1/services/ hybrid_search.py, business_logic.py
├── module2/services/ + agents/ agent_tools.py, recommendation_agent.py, orchestrator.py
└── module3/services/ + frontend/ agentcore_*.py, otel_trace_extractor.py, agentIdentity.ts
scripts/ # Bootstrap & provisioning ├── bootstrap-environment.sh Stage 1: Code Editor + Python setup ├── bootstrap-labs.sh Stage 2: DB seed + deps (workshop) ├── bootstrap-labs-builders.sh Stage 2: Same + pre-complete C3-C9 (builders) └── seed-database.sh Product catalog + return_policies + indexes
data/ # Product catalog (~444 products with embeddings) lab-content/ # Workshop Studio content (2 formats) ├── workshop/ 2-hour workshop lab guide └── builders/ 1-hour builder's session lab guide
---
## Technology Stack
| Layer | Technologies |
|-------|-------------|
| Database | Aurora PostgreSQL Serverless v2 (0-16 ACU), pgvector 0.8.0 (HNSW) |
| AI/ML | Amazon Bedrock — Claude Sonnet 4.6, Cohere Embed v4, Cohere Rerank v3.5 |
| Agent Infra | Amazon Bedrock AgentCore — Gateway, Memory, Observability, Runtime |
| Agent Framework | Strands Agents SDK |
| Backend | FastAPI, Python 3.13, psycopg3, boto3 |
| Frontend | React 18, TypeScript 5, Tailwind CSS, Vite |
---
## Multi-Agent Architecture
5 specialist agents + 1 orchestrator, using the "Agents as Tools" pattern:
| Agent | Domain | Tools |
|-------|--------|-------|
| Search | Product search | `search_products`, `get_product_by_category`, `compare_products` |
| Recommendation | Trending/popular | `get_trending_products`, `get_product_by_category` |
| Pricing | Price analysis | `get_price_analysis`, `search_products`, `get_product_by_category` |
| Inventory | Stock management | `get_inventory_health`, `get_low_stock_products`, `restock_product` |
| Support | Returns/support | `get_return_policy`, `search_products` |
| Orchestrator | Query routing | The 5 specialist agents (registered as tools) |
---
## Short on Time?
Every challenge has a solution file. Copy it over and the backend auto-restarts:
```bash
# Example: Skip Challenge 1
cp solutions/module1/services/hybrid_search.py blaize-bazaar/backend/services/hybrid_search.py
See solutions/README.md for all copy commands.
Table: blaize_bazaar.product_catalog (~444 products)
CREATE TABLE blaize_bazaar.product_catalog (
"productId" CHAR(10) PRIMARY KEY,
product_description VARCHAR(500) NOT NULL,
"imgUrl" VARCHAR(200),
"productURL" VARCHAR(40),
stars NUMERIC(2,1),
reviews INTEGER,
price NUMERIC(8,2),
category_id SMALLINT,
"isBestSeller" BOOLEAN DEFAULT FALSE,
"boughtInLastMonth" INTEGER,
category_name VARCHAR(50) NOT NULL,
quantity SMALLINT,
embedding vector(1024) -- Cohere Embed v4
);Table: blaize_bazaar.return_policies (21 rows)
CREATE TABLE blaize_bazaar.return_policies (
category_name VARCHAR(50) PRIMARY KEY,
return_window_days INTEGER,
conditions TEXT,
refund_method TEXT
);- Aurora PostgreSQL with pgvector
- Amazon Bedrock AgentCore
- Strands Agents SDK
- pgvector 0.8.0 Performance
- AWS Labs MCP Servers
MIT-0 License. See LICENSE.
Workshop by Shayon Sanyal — Principal Database Specialist SA, AWS