Revenue-ops vertical · live in production

GTM-OS: founder go-to-market operating system on SoloFrame.

An AI sales coaching platform for solo founders: 49-course / 487-lesson curriculum localized across English, Spanish, and Brazilian Portuguese surfaces (1,461 lesson files total), founder-tuned RAG coaching, AI roleplay sales training across 11 DISC-typed LatAm buyer personas, pod-based community matching, a 21-chapter book product, and an API-integrations layer. LatAm is the first GTM-OS localization proof, not the ceiling. The non-PHI commercial counterpart in the platform's manifest contrast — same engine as DWA, opposite compliance posture, completely different domain.

Cost-architecture flagship

6.2x measured per-session cost reduction.

A 20-turn founder coaching session: $0.594 naive (frontier-only Sonnet) - $0.0955 SoloFrame. The headline number for SoloFrame's six-layer cost architecture anchors here at GTM-OS, not at DWA. DWA's 4.4x is the secondary data point (shorter sessions amortize fewer turns); GTM's 6.2x is the load-bearing claim. Reproducible from docs/positioning/unit-economics-model.csv using vendor-published pricing.

6.2x
GTM measured · current
4.4x
DWA measured · secondary
6.7x
Layer 6 conservative
13.6x
Layer 6 ambitious
LayerWhat it isWhere in GTM
1
Tier routing
16+ AI tasks each map to env override + fallback: moonshotai/kimi-k2 for coaching, roleplay, outreach drafts; deepseek/deepseek-v3.2 for reasoning-heavy eval, pipeline-insights, outreach-analysis; google/gemini-2.5-flash for long-context extraction; qwen/qwen3-235b-a22b for open-weight defaults; google/gemini-2.5-pro for assessment scoring
lib/ai/models.ts
2
OpenRouter model bus
Single OpenAI-SDK client pointed at OpenRouter. One bill, many models. Voice (Whisper STT + tts-1) bypasses to OpenAI direct
@svtech/llm + lib/ai/client.ts
3
Prompt + context caching
Anthropic-style cache breakpoints on stable prefixes. ~90% read discount on cached portion within 5-min TTL
lib/ai/openai-coaching.ts
4
Two-corpus RRF RAG
Reciprocal-rank-fusion over platform corpus (487 lessons, 4,998 chunks) + founder customer corpus (uploaded PDFs). RRF_K=60 damping. Invocation regex shifts weighting on 'use my upload'
lib/services/ragService.ts
5
CPU sidecar classifiers
Not wired in GTM. classifier: 'noop' selects defaultNoOpClassifier - founder coaching frustration is not clinical distress
manifest.coaching.classifier
6
In-house fine-tunes
Deferred. When live, light turns shift from qwen/kimi/deepseek to self-hosted coach. Conservative 6.7x; ambitious 13.6x with 60% of heavy reasoning absorbed
services/maia/training/ + Nebius (pending)
Curriculum · 49 courses · 7 tracks

1,461 localized lesson files across English, Spanish, and Brazilian Portuguese surfaces.

The Spanish-language corpus is one artifact of the localization system, not the whole claim. Locale support is a manifest knob (locale.supported: ["en", "es"]), while the first LatAm GTM-OS localization also carries localized buyer personas, country tone, regional workflows, and integration choices without a platform fork. The point is repeatability: future GTM-OS markets should follow the same pattern.

TrackCoursesEN lessonsES lessons
foundationssales-psychology · icp-builder · positioning-value · list-building · choose-path4848
marketing-enginecold-email-mastery · linkedin-engine · email-nurture · seo-aeo · technical-content · social-proof-referral · marketing-automation-analytics · community-lead-gen8888
sales-methodologydiscovery-framework · discovery-simulations · demo-architecture · proposals-pricing · objection-handling · closing-closing · pipeline-management · disc-personas8383
ai-acquisitionai-acquisition-strategy · email-deliverability · ai-lead-research · ai-outreach-automation · linkedin-ai · autonomous-sdr · custom-ai-agents7676
creator-trackcreator-sales-mindset · audience-to-buyer · webinar-challenge-funnels · dm-selling-social-commerce · creator-sales-conversations · community-led-sales · scaling-creator-sales · creator-metrics7272
customer-successonboarding · retention · expansion · advocacy3636
operations-systemscrm-setup · analytics · automation · playbook · scale · outsourcing · legal · finance · capstone8484
Total49 courses487487
Roleplay engine

11 DISC-typed buyer personas. 5 country variants. Practice infrastructure, not a chatbot.

Where DWA has MAIA-driven safety routing, GTM-OS has a roleplay-driven practice loop with structured post-session scoring. The structured score is the data-flywheel claim made empirical for GTM in the same way MAIA's classification-event signal is for DWA.

11 named personas

6 LatAm Spanish + 5 English

Spanish (es): Alejandro VP Comercial (D-type, Bogotá) · Camila Fundadora EdTech (I-type) · Sebastián CFO Healthtech (C-type, formal "usted") · Valentina Directora Ventas (D-type) · Andrés Director Marketing (I-type) · Isabella Gerente Compras (C-type, procurement)
English: Technical Skeptic CTO (C-type) · Visionary Founder (I-type) · Procurement Blocker (C-type) · Stressed VP Sales (D-type) · Marketing Juggler (I/S-type)

Each persona carries DISC type, register (tu/usted), city, country, personality, common objections, hidden agenda, difficulty. Defined at lib/data/personas.ts (500 LOC).

5 country variants

LatAm-localized cultural notes

Colombia (default): respect-and-hierarchy patterns, Bogotá-default tone
Mexico: Mexican formality cues
Chile: distinctive soft-no patterns
Argentina: confianza-tone, distinct from Colombian respect register
Spain: European-Spanish register, currency cues

Each variant injects a cultural note, tone description, country-specific soft-no patterns, currency/formality cues into the system prompt at session start.

Roleplay session lifecycle
1Industry + role select
Founder picks industry (8 LatAm-shaped) + clientRole. DISC type encoded via _high_<DISC> suffix
2Context build
buildRoleplayContext assembles founder profile, archetype, industry, DISC pattern, country variant, locale
3System prompt build
buildRoleplaySystemPrompt composes persona + DISC + country tone + register + cultural notes + objection library
4Streaming conversation
api/ai/roleplay - openaiRoleplayReply streams persona responses (Kimi K2 default). Per-turn rate-limited
5Post-session evaluation
api/ai/roleplay/evaluate runs DeepSeek V3.2 for reasoning-heavy critique. Structured score across MVQ-Framework dimensions
6Certification gate
Average roleplay score >= 75 is one of three Certified Solo GTM Practitioner criteria
8 guided AI workshops

Multi-step AI-guided artifact builders. Edit-in-place, AI-reviewed, persisted.

Each workshop has an assessmentDimension, framework reference, estimated minutes, output artifact, and step types: show-current, teach-framework, guided-edit, ai-review, save-artifact. Some trigger a re-audit (LinkedIn, website, documents) after completion.

ICP clarity
ICP Framework · 25 min
Outputs: icpDocument
Positioning rewrite
Prescription Frame · 30 min
Outputs: positioningDocument
LinkedIn overhaul
Profile rewrite · 45 min
Outputs: linkedinProfile
re-audits source after
Channel selection
Playbook Selection · 20 min
Outputs: channelChoice
Discovery script
MVQ Framework · 35 min
Outputs: discoveryPlaybook
Sprint cold email
Zero-to-Ten · 40 min
Outputs: emailSequence
Sprint LinkedIn profile
Zero-to-Ten · 30 min
Outputs: linkedinProfile
Sprint sales script
Zero-to-Ten · 35 min
Outputs: salesScript
Execution layer · May 2026

Five AI components that close the Learn→Execute loop.

Pre-May 2026 GTM had strong learning and artifact-building but no path from built artifacts to executable outreach, and no feedback from execution results back into coaching. The execution layer closes both gaps: artifacts in → content out, pipeline data in → coaching context out.

Outreach Draft Generator
api/ai/generate-drafts
model: moonshotai/kimi-k2

ICP + positioning + email sequences → 5 cold email drafts (tone-labeled) + LinkedIn note + cold-call opener. Two-pass: generate at temp 0.9, refine at temp 0.2 (subject under 50 chars, opens with buyer pain, CTA is one ask). Full LatAm cultural localization for CO/MX/CL/AR.

Pre-Call Brief
api/ai/call-brief
model: qwen/qwen3-235b-a22b

Prospect name/company/role → ICP fit badge, recommended opening angle, discovery questions, key value points, likely objections with responses. Reads icpDocument, discoveryPlaybook, objectionLibrary, valuePropositionCanvas from profile.

Pipeline Intelligence
api/ai/pipeline-insights
model: deepseek/deepseek-v3.2

Live pipeline_deal + outreach_log DB data → health score + typed bottleneck/opportunity/warning/coaching insights. Stale-deal detection with per-stage thresholds. Renders as PipelineCoachWidget above the kanban board.

Outreach Analysis
api/ai/outreach-analysis
model: deepseek/deepseek-v3.2

30-day outreach_log → reply-rate diagnosis, winning pattern, suggested rewrites. Requires ≥3 logged activities. Returns structured JSON with example rewrites the founder can copy-paste.

Artifact Outcomes Tracker
api/artifacts/[type]/outcomes
model:

Founders log sends/replies/meetings per artifact version. liveOutcomes string injected into every coaching context — the coach references real performance data to diagnose by reply-rate thresholds, not generic advice.

Coaching memory loop. Session embeddings stored in profile.coachingSessions. Each coaching turn embeds the current message and retrieves the 3 most semantically relevant past sessions (cosine similarity) instead of always injecting the last 3. Falls back to recency for sessions without embeddings. The coach's context window reflects what's actually relevant, not just what's recent.
API integrations

Eight integration clients. CRM, outreach, enrichment, community, badges.

A solo founder threads their own CRM / outreach / enrichment / community / badge tools through GTM-OS without GTM having to onboard them as licensees. Per-founder credentials encrypted via @svtech/security; the connectedAccount table is namespaced by provider + scope with per-account rate limits, retry/backoff, and credential rotation surfaces.

ProviderAuth patternWhat it does
AttioPlatform-side singleton + retryFounder records sync to Attio CRM as deals progress
Hunter.ioFounder-supplied API keyDomain-search enrichment for cold outreach
PipedriveFounder-supplied API keyLatAm-popular CRM (Brazil, Mexico, Colombia, Argentina). Deal sync surface
Notion3-legged OAuthFounder uses their Notion workspace for artifact storage
WhatsApp BusinessFounder Meta Business SuiteEssential for LatAm market - 62-80% of users communicate with businesses via WhatsApp. 12-week-sprint message templates
Brevo (ex-Sendinblue)Founder-supplied API keyDominant email marketing in LatAm. Transactional + campaigns + contacts + lists
BadgrPlatform-side env varOpen Badges 3.0 issuance. Awards Certified Solo GTM Practitioner badge
NodeBBMaster-token API · graduatedThe only adapter promoted to a @svtech/* package today (@svtech/adapter-nodebb)
Community · pods + forum bots

4-6 founder cohorts, matched by hard-then-soft constraints. 5 personality-tagged AI bots seed cold-start activity.

The matching algorithm lives at lib/services/podMatchingService.ts (389 LOC). 5 forum bots create the first-mover seed activity that solves the cold-start problem of a new community surface.

Pod matching algorithm

Hard constraints
  • Curriculum stage compatibility (foundation / lead-gen / sales-conversation; ±1 stage allowed)
  • Time-commitment compatibility (5-10h, 10-15h, 15-20h, 20h+; adjacent buckets)
Soft scoring
  • DISC diversity (avoid all-D-type or all-I-type pods)
  • Deal-size clustering (similar ARR / contract sizes)
  • Industry diversity (variety preferred)
  • Pod-health metrics (existing pod activity, completion rates)

5 forum bots seed cold-start activity

  • facilitator-bot"AI Pod Coach" - surfaces stuck threads, weekly summaries, flags quiet members
  • alex-skeptic-botD-type "Skeptical Challenger" - pushes back on under-baked plans
  • jordan-builder-botI-type "Relationship Builder" - encourages connection, celebrates wins
  • morgan-perfectionist-botC-type - challenges precision, asks for data
  • sam-mentor-botS-type - steady mentorship voice, reframes setbacks

Replies via api/internal/persona-respond using the facilitator (Kimi K2) and persona (Qwen) tasks.

Credentials + content products

Certified Solo GTM Practitioner pathway. Founder and Small Team GTM.

Certification

Certified Solo GTM Practitioner

Three eligibility criteria, all required:

  1. All Track 1 (Foundations) courses complete
  2. All Track 3 (Sales Methodology) courses complete
  3. Roleplay average score >= 75 across completed sessions

On award: profile.progress.certificationEarned set (idempotent), Badgr Open Badges 3.0 assertion issued (locale-aware EN/ES badge IDs), founder surfaces on the public /certified directory, Resend transactional email triggered.

Book product

"Founder and Small Team GTM"

By Mike Sullivan. 33 .md files in docs/manuscript/ covering front matter, 15 chapters, parts, back matter. Reading surface at /book + per-chapter reader.

  • book_purchase via Polar (separate product ID)
  • book_search_index full-text search built at boot
  • book_reading_event tracking; "you've read X%" gauge on dashboard
  • BOOK_AUTHOR_PERSON JSON-LD with full Person schema for E-E-A-T density

The non-PHI side of the manifest contrast.

GTM-OS is the live receipt that the engine handles a non-clinical workload without the regulatory tax DWA carries. Same Mono-PaaS engine, opposite compliance posture, completely different domain - and the 6.2x cost-arch headline.