7. Team Size — Building and Operating MangaAssist

Team Structure Overview

graph TD
    subgraph "Leadership"
        PM[Product Manager<br>1 person]
        TPM[Technical Program Manager<br>1 person]
    end

    subgraph "Engineering"
        SDE[Senior Software Engineers<br>2-3 people]
        FE[Frontend Engineers<br>2 people]
        BE[Backend Engineers<br>3-4 people]
        ML[ML / Applied Scientists<br>2 people]
    end

    subgraph "Platform & Quality"
        DEVOPS[DevOps / MLOps<br>1-2 people]
        DATA[Data Engineers<br>1-2 people]
        QA[QA / SDET<br>2 people]
    end

    subgraph "Design & Ops"
        UX[UX Designer<br>1 person]
        OPS[Support / Ops<br>1-2 people]
    end

    PM --> SDE
    TPM --> SDE
    SDE --> FE
    SDE --> BE
    SDE --> ML
    BE --> DEVOPS
    ML --> DATA

Role Breakdown

Product Manager (1 person)

Aspect	Detail
Why needed	Defines what the chatbot should do, prioritizes features, owns the roadmap
What they do	Writes requirements, defines success metrics, coordinates with business stakeholders (JP Manga store team), runs A/B experiments
MVP	1 PM
Full rollout	1 PM (same person, more stakeholder management)

Technical Program Manager (1 person)

Aspect	Detail
Why needed	This project touches 6+ internal teams (catalog, orders, recommendations, support, ML platform, frontend platform). Someone must coordinate dependencies and timelines.
What they do	Tracks cross-team dependencies, runs sprint ceremonies, manages launch readiness, escalates blockers
MVP	0-1 (Senior Dev can cover in MVP)
Full rollout	1 TPM

Senior Software Engineers / Senior Developers (2–3 people)

Aspect	Detail
Why needed	Own the architecture, make technical decisions, mentor the team, write the hardest code
What they do	Design the orchestrator, define API contracts, build the RAG pipeline, own scalability and reliability, review all PRs, drive operational excellence
MVP	2 (one owns backend/orchestration, one owns AI/RAG)
Full rollout	3 (add one for platform/infrastructure)

Frontend Engineers (2 people)

Aspect	Detail
Why needed	The chat widget must integrate cleanly into Amazon.com's existing frontend, handle streaming, and work across web and mobile
What they do	Build the React chat widget, implement WebSocket streaming, build product card rendering, handle accessibility, optimize bundle size
MVP	1
Full rollout	2 (add mobile-specific work)

Backend Engineers (3–4 people)

Aspect	Detail
Why needed	Multiple services to build: orchestrator, intent router, conversation memory, integration adapters for catalog/orders/recommendations
What they do	Implement service logic, write integration clients, build the guardrails pipeline, handle error cases, implement rate limiting
MVP	2
Full rollout	3-4

ML / Applied Scientists (2 people)

Aspect	Detail
Why needed	Intent classifier needs training data and model tuning. RAG pipeline needs embedding selection, chunk strategy optimization, and reranking. Prompt engineering is iterative and needs experimentation.
What they do	Train and deploy the intent classifier, optimize RAG retrieval quality, tune LLM prompts, build evaluation datasets, measure hallucination rates
MVP	1 (focused on prompt engineering + RAG)
Full rollout	2 (add dedicated intent classifier + recommendation integration work)

DevOps / MLOps / Platform Engineers (1–2 people)

Aspect	Detail
Why needed	CI/CD pipelines, model deployment, infrastructure-as-code, monitoring setup
What they do	Set up CDK/CloudFormation, configure Bedrock endpoints, build deployment pipelines, set up CloudWatch dashboards, manage SageMaker endpoints
MVP	1
Full rollout	2

Data Engineers / Analytics Engineers (1–2 people)

Aspect	Detail
Why needed	Analytics pipeline (Kinesis → Redshift), dashboards, RAG indexing pipeline, data quality
What they do	Build the event pipeline, create Redshift tables and dashboards, maintain the RAG indexing job, ensure data freshness
MVP	1
Full rollout	2

QA / SDET (2 people)

Aspect	Detail
Why needed	Chatbot responses are non-deterministic. Traditional testing isn't enough — need evaluation frameworks, regression test suites for intents, and E2E testing.
What they do	Build automated test suites, create intent classification test sets, test guardrails, run load tests, validate integration points
MVP	1
Full rollout	2

UX Designer (1 person)

Aspect	Detail
Why needed	Chat UX is nuanced — message bubbles, product cards, quick chips, streaming indicators, error states, mobile layout all need careful design
What they do	Design the chat widget, create interaction patterns, user-test prototypes, define the design system for chatbot responses
MVP	1
Full rollout	1

Support / Operations (1–2 people)

Aspect	Detail
Why needed	Someone needs to monitor chatbot quality daily, review escalated conversations, update the knowledge base, and handle incidents
What they do	Daily quality review, KB updates, incident response, feedback triage, coordinate with Amazon CS team for escalation flow
MVP	0 (engineering covers)
Full rollout	1-2

Team Size Summary

Phase	Total Headcount	Duration
MVP	10–12 people	3–4 months
V2 (Post-purchase + personalization)	15–18 people	3 months additional
Full Production	18–22 people	Ongoing
Steady State (maintenance)	8–12 people	Ongoing