Overview·Topics·Comparison·Next Steps

BUILD SPRINT

TOPIC PROPOSALS

Berlin — June 2026

Overview·Topics·Comparison·Next Steps

OVERVIEW

WHAT WE'RE DECIDING

Overview·Topics·Comparison·Next Steps

WHAT IS THE BUILD SPRINT?

  • A dedicated sprint to invest in Gemma's own capabilities — new services, internal tools, and reusable assets
  • Each topic has a clear Definition of Done, ROI hypothesis, and time cap
  • We're selecting which topics to commit to for the next sprint

Goal: Pick the right topics, assign ownership, and walk out with a shared commitment.

Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

8 PROPOSALS AT A GLANCE

8
Total proposals
2
New services
4
Internal tools
1
Prototype
1
Best practice
Revenue-focused: Topics #31, #32, #33, #37
Efficiency-focused: Topics #35, #36, #38, #39
Time range: 8h – 20h per topic
Effort: Low to Medium
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

TOPICS

THE EIGHT PROPOSALS IN DETAIL

Overview·Topics·Comparison·Next Steps

#31 — CONVERSATIONAL ANALYTICS

BEDROCK + METABASE MCP

  • Business users wait for analysts to answer ad-hoc data questions — creating a bottleneck
  • LLMs + Metabase MCP now make natural language analytics viable
  • Build a conversational layer that sits in front of Metabase and answers questions in plain language
ROI: Saves 3–5h per analyst per week. Enables a new repeatable service offering positionable in 3–5 client proposals by Q3 2026.

Category: New Service
Owner: Elena
Effort: Normal
Time Cap: —
ROI Type: Revenue
Priority: —

Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#31 — DEFINITION OF DONE

Working conversational analytics agent
Deployable as a scoped prototype, demoable to a client stakeholder
Reusable agent template
Documented and generalized — deployable against any client Metabase with minimal config
Bedrock + Metabase MCP blueprint
Written and version-controlled — technical foundation for future engagements
Handable to delivery team
A Gemma delivery team can onboard a client without the builder present
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#32 — OPEN-SOURCE LLM EVALUATION

EU SOVEREIGNTY PLAYBOOK

  • Regulated EU prospects cannot use US-hosted LLM APIs (DSGVO, data residency mandates)
  • Gemma currently has no structured, evidence-backed answer to this objection in sales conversations
  • Evaluate open-source LLMs on a self-hosted EU VM and produce a sales-ready playbook
ROI: Convert EU sovereignty-sensitive prospects by giving sales leads a tested, credible answer to the most common blocker — measurable by Q4 2026.

Category: Prototype
Owner: —
Effort: Medium
Time Cap: 16h
ROI Type: Revenue
Priority: High

Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#32 — DEFINITION OF DONE

2+ LLMs evaluated on EU VM
Setup, latency, cost, and quality benchmarked on a self-hosted instance
Mini playbook
Model picks, hosting overview, cost vs. proprietary APIs
Reusable harness
Ready to deploy for client projects without rebuilding from scratch
Case study written
EU sovereignty, DSGVO, US-provider independence — test-based
1 sales-ready summary slide
Immediately usable in client-facing proposals
Signed off
By 1 sales lead + 1 engineer
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#33 — "MAKE THE ORG AI" SERVICE

PACKAGED AI ADOPTION SERVICE

  • Most organisations want to adopt AI but don't know where to start — the gap is human, not technical
  • Gemma is ahead internally but that knowledge is unstructured and not packaged for delivery
  • Time-sensitive: package this before it becomes a commodity
ROI: New repeatable revenue stream — first service package sold or piloted by Q4 2026. Converts AI-curious prospects into paying clients.
💬 Open question: Still relevant after Elena worked on the proposal for Argos?

Category: New Service
Owner: Elena
Effort: Low
Time Cap: 20h
ROI Type: Revenue
Priority: Normal

Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#33 — DEFINITION OF DONE

Internal rollout documented
Which tools, how, and by whom — baseline for client package
MCP server setup guide
Validated by a Gemma engineer who didn't build it
Claude / Claude Code onboarding flow
Account setup, first use cases, guardrails
Workshop format designed
Agenda, exercises, facilitator notes for a half-day client session
Change management playbook
Max 5 pages: phases, roles, adoption milestones, client leave-behind
Service package scoped
Clear offering, indicative pricing, delivery format — signed off by sales + delivery
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#35 — SEARCHABLE KNOWLEDGE BASE

  • Internal knowledge is fragmented across Confluence, GitHub, Slack, and dbt docs
  • Finding the right answer relies on tribal knowledge — new team members are hit hardest
  • Build a chat-based AI interface that searches across all sources and returns relevant, up-to-date answers
ROI: Saves every team member time weekly by replacing multi-tool searching with a single AI entry point — measurable from month 1 via team survey.
💬 Merge together with #38 Pure-Git Knowledge Base — Confluence Replacement Pilot

Category: Internal Tool
Owner: Matthew
Effort: Medium
Time Cap: 16h
ROI Type: Efficiency
Priority: Normal

Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#35 — DEFINITION OF DONE

Chat interface live
Natural language questions return relevant answers from at least Confluence and GitHub
Source links included
Every answer links back to the original source document
Staleness rules defined
Outdated content filtered out automatically or flagged clearly
Solution documented
Setup + usage guide for ongoing maintenance
Piloted internally
Positive feedback from at least 3 key stakeholders
Quality validated
10+ representative test questions (printer setup, Permifrost, dbt best practices)
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#38 — PURE-GIT KNOWLEDGE BASE

CONFLUENCE REPLACEMENT PILOT

  • Confluence is slow, expensive, hard to version, and not designed for LLM interaction
  • A pure-git, text-file-first approach (.md + .json in GitHub) makes knowledge natively usable by Claude, version-controlled, and vendor-lock-in free
  • Pilot: replace Confluence for one team or use case to test viability before a full migration
ROI: Saves ongoing Confluence cost and friction — every page added becomes immediately usable by Claude, the sales website framework, and any future AI workflow.

Category: Internal Tool
Owner: Bijan
Effort: High
Time Cap: —
ROI Type: Efficiency
Priority: Normal

Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#38 — DEFINITION OF DONE

Repo structure defined
Folder conventions, .md schema for pages, .json schema for database-like entries
Lightweight web UI
Non-technical team members can browse and read content without GitHub access
Rights management
Read/write separation between roles — CODEOWNERS or branch protection
Real content migrated
At least one Confluence section moved into the pilot repo
LLM-readable validated
Claude can answer questions from repo content without preprocessing
2-week pilot with 3+ users
Go/no-go recommendation written for leadership
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#36 — HTML-FIRST PRESENTATIONS REPO

BRANDED, VERSION-CONTROLLED DECKS

  • No consistent, branded way to produce AI-generated presentations
  • Ad hoc approaches lead to inconsistent quality, off-brand visuals, and no version control
  • Build an HTML-first repo with Gemma branding, reusable templates, and a Claude skill for slide generation
ROI: Saves 1–2 hours per presentation per consultant. Raises the quality floor of every client deliverable from day one.

Category: Internal Tool
Owner: Bijan
Effort: Medium
Time Cap: 12h
ROI Type: Efficiency
Priority: Normal

Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#36 — DEFINITION OF DONE

Framework decision documented
e.g. Reveal.js, Slidev, or custom — with rationale
Repo with clear structure
Templates, assets, examples, and skill(s)
Gemma brand assets integrated
Colours, fonts, logo, spacing system
3+ reusable slide layouts
Title, content, data/chart slides
2+ worked examples
Real Gemma output types (proposal, results readout)
Validated by non-builder
Consultant with no HTML knowledge produces a deck in under 1 hour
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#37 — PERSONALISED SALES WEBSITES

AI-GENERATED MICROSITES PER LEAD

  • PDF proposals are static, generic, and invisible once sent — no engagement signal, no personalisation
  • Building a simple website per lead is now a 1–2 day problem with AI
  • Replace PDFs with personalised microsites: considered, easy to navigate, generated from a brief in minutes
ROI: More pipeline by replacing forgettable PDF proposals with personalised, shareable microsites — first live prospect site within 4 weeks.

Category: Internal Tool
Owner: Bijan
Effort: Medium
Time Cap: 12h
ROI Type: Revenue
Priority: Normal

Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#37 — DEFINITION OF DONE

Hosting approach decided
Documented — shareable URL per lead
Base template with blocks
Personalised value prop, selectable case studies, clear CTA
Case studies structured centrally
Any subset includable per site without content duplication
AI generation flow
Consultant provides brief → AI drafts → human reviews and publishes
Non-technical consultant test
Brief to shareable URL in under 2 hours, no HTML experience needed
Worked example + sign-off
1 realistic anonymised prospect site, validated by 1 sales lead
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#39 — DBT AI READINESS FRAMEWORK

MAKING DBT PROJECTS LLM-READY

  • LLMs are being plugged into dbt projects via MCP, Metabase AI, and Claude — but the projects weren't built for machine readers
  • Without structured metadata, semantic definitions, and explicit lineage, every AI tool guesses or fails
  • Build a reusable, opinionated AI readiness framework for dbt as a Gemma delivery standard
ROI: Reduces per-project AI integration risk and generates a repeatable, sellable capability for clients connecting LLMs to their data warehouse safely.
💬 Merge together with #31 Conversational Analytics

Category: Best Practice
Owner: Bianca (idea: David)
Effort: Medium
Time Cap: 12h
ROI Type: Efficiency
Priority: High

Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

#39 — DEFINITION OF DONE

meta: fields standard
Model-level: owner, domain, certified, grain, contains_pii, synonyms. Column-level: contains_pii
Exposures framework
Structure defined, at least one real exposure per output type, forward lineage traceable
Semantic Layer research
MetricFlow: cost, dbt Cloud dependency, maturity, go/no-go recommendation
dbt Governance assessed
Groups, Access, Contracts — each with adopt/defer recommendation
Proof-of-concept applied
Applied to at least one real client dbt project, validated with LLM
AI readiness checklist packaged
Max 1 page, validated by 1 delivery consultant + 1 engineer
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

COMPARISON

SIDE BY SIDE

Overview·Topics·Comparison·Next Steps

TOPIC COMPARISON

# Topic Category ROI Type Time Cap Effort Owner Priority
31 Conversational Analytics New Service Revenue — Normal Elena —
32 Open-Source LLM Eval Prototype Revenue 16h Medium — High
33 "Make the Org AI" New Service Revenue 20h Low Elena Normal
35 Knowledge Base Internal Tool Efficiency 16h Medium Matthew Normal
38 Pure-Git Knowledge Base Internal Tool Efficiency — High Bijan Normal
36 HTML Presentations Internal Tool Efficiency 12h Medium Bijan Normal
37 Sales Websites Internal Tool Revenue 12h Medium Bijan Normal
39 dbt AI Readiness Best Practice Efficiency 12h Medium Bianca High
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

ROI LENS

💰 Revenue-generating (4 topics)
  • #31 Conversational Analytics — new repeatable service module
  • #32 Open-Source LLM — unblock DSGVO-sensitive deals
  • #33 "Make the Org AI" — new service package for AI adoption
  • #37 Sales Websites — more pipeline via personalised microsites
⚡ Efficiency-boosting (4 topics)
  • #35 Knowledge Base — eliminate multi-tool searching internally
  • #38 Pure-Git KB — replace Confluence, make knowledge LLM-native
  • #36 HTML Presentations — save 1–2h per deck, raise quality floor
  • #39 dbt AI Readiness — make dbt projects safe for LLM integration
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

TIME INVESTMENT

#32 OS LLM Eval
16h
HIGH PRIORITY
#36 HTML Presentations
12h
#37 Sales Websites
12h
#39 dbt AI Readiness
12h
HIGH PRIORITY
#35 Knowledge Base
16h
#33 "Make the Org AI"
20h
#31 Conv. Analytics
TBD
#38 Pure-Git KB
TBD
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

OWNERSHIP MAP

ELENA
#31 Conv. Analytics
#33 "Make the Org AI"
2 topics · Revenue
BIJAN
#36 HTML Presentations
#37 Sales Websites
#38 Pure-Git KB
3 topics · Mixed
MATTHEW
#35 Knowledge Base


1 topic · Efficiency
BIANCA
#39 dbt AI Readiness


1 topic · Efficiency
⚠️ #32 Open-Source LLM Eval has no owner assigned yet — highest priority topic
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

NEXT STEPS

DECISIONS TO MAKE

Overview·Topics·Comparison·Next Steps

DECISIONS FOR TODAY

1
Which topics do we commit to?
We have 8 proposals — do we take all of them, or prioritise a subset?
2
Who owns #32 (Open-Source LLM)?
It's marked High priority with an 8h cap but has no owner. Who picks it up?
3
What's the time cap for #31?
Conversational Analytics has no time cap yet — do we scope it before committing?
4
Sprint timeline and check-in cadence
When does the sprint start, when do we check in, and when is the demo?
Gemma Analytics | BUILD Sprint — Topic Proposals
Overview·Topics·Comparison·Next Steps

LET'S DISCUSS

WHICH TOPICS DO WE BUILD?