TEGI — PM Analysis

1. Project Charter

Project Definition

Project Name

TEGI — The Entity Graph Internet

Vision

Build the canonical identity, knowledge, and agent interface layer for every entity on the internet.

North Star

"The web connects pages. TEGI connects entities."

Team

Dedicated technical team — full-stack TypeScript, ML/NLP infrastructure, and distributed systems expertise

Build Window

12-month roadmap to Phase 5 (enterprise billing). MVP targeted at Month 6.

Primary Market

B2B first: companies, products, institutions, AI agent deployments

Stack

TypeScript monorepo (Next.js 16, tRPC, Postgres) + Python microservice (File Clerk)

Definition of Done

type-check → lint → test → build → test:e2e:stable — all passing

2. 12-Month Roadmap

Phase Timeline

Phase

M10

M11

M12

Phase 1Identity

Identity & Trust

Phase 2Knowledge

File Clerk + RAG

Phase 3Entity Graph

Graph + D3 UI

⭐ Phase 4MVP

⭐ MVP

Agent Runtime

Phase 5Transactions

Billing + Enterprise

Phase 6Autonomous

Dev API + LoRA

Phase 1 — Identity

Phase 2 — Knowledge

Phase 3 — Graph

Phase 4 — MVP ⭐

Phase 5 — Transactions

Phase 6 — Autonomous

Timeline	Phase	Deliverables
M1–2	Phase 1 — Identity	Entity table, trust tiers, owner auth, basic feed, entity profiles, unified authorship
M2–3	Phase 2 — Knowledge	File Clerk Python microservice, knowledge_items, pgvector, Qdrant, semantic search, RAG retrieval
M3–4	Phase 3 — Graph	entity_edges table, typed edge CRUD, relationship browsing, graph-aware search, D3 graph UI
M4–6 ⭐	Phase 4 — MVP	Entity-attached agents, reactive replies, direct sessions, context profiles, transparency layer
M6–9	Phase 5 — Transactions	Action framework, token billing, token margin infrastructure, enterprise onboarding
M9–12	Phase 6 — Autonomous	Auto graph enrichment, reputation engine, developer API, agent marketplace, LoRA pipeline

3. Task Breakdown — Phases 1–4

Engineering Task List

Phase 1 — Identity (M1–2)

T01

Postgres schema — entities table, trust_tier enum, owner_id FK

T02

NextAuth setup — GitHub, LinkedIn, Google OAuth + magic link

T03

Entity CRUD — create, claim, update, soft-delete tRPC procedures + IDOR tests

T04

Trust tier gates — feature flags by tier, admin override, dispute flow

T05

Basic feed — post creation, reactions, comments, pagination

T06

Entity profiles — public view, claimed view, edit mode, trust badge display

T07

AI disclosure — is_ai_agent enforcement, banner on all agent entity surfaces

Phase 2 — Knowledge (M2–3)

T08

File Clerk microservice scaffold — FastAPI, BullMQ consumer, Docker service

T09

Document parsing — unstructured integration, PDF/HTML/DOCX/MD support

T10

Chunking + embedding — sentence-transformers, configurable chunk size

T11

Vector storage — pgvector index, Qdrant collection, dual write

T12

RAG retrieval — hybrid dense + BM25, tRPC endpoint for knowledge query

T13

knowledge_items table — schema, CRUD, entity linkage, source provenance

Phase 3 — Graph (M3–4)

T14

entity_edges table — typed edges, weight, metadata, bidirectional index

T15

Edge CRUD — create, update, delete tRPC procedures + IDOR tests

T16

Graph-aware search — rank results by relationship proximity

T17

Relationship browsing UI — entity profile graph tab

T18

D3 graph explorer — force-directed layout, filter by edge type and trust tier

Phase 4 — MVP (M4–6)

T19

entity_agents table — model, system_prompt, allowed_actions, tools config

T20

Agent runtime — multi-model routing, Claude/GPT/Gemini adapter layer

T21

Reactive agent replies — agent auto-responds to forum threads and feed mentions

T22

Direct session UI — private chat, session persistence, full history

T23

Hot context — last N turns injected into session prompt

T24

Cold context — RAG over interaction history, Qdrant nearest-neighbour

T25

Domain profile — passive interest inference, category scoring

T26

Transparency layer — 'What do you know about me?' button, footprint page, GDPR delete

4. Risk Register

Identified Risks & Mitigations

ID	Risk	Prob	Impact	Mitigation
R01	Scope creep / milestone drift	Med	Medium	Strict phase gating; weekly milestone reviews; MVP scope locked before Phase 4 kick-off
R02	File Clerk latency	Med	Medium	Async queue via BullMQ; user-facing progress indicator; background processing
R03	LLM cost overrun	Med	High	Token margin covers cost; per-entity rate limits; model routing to cheapest capable model
R04	Trust tier fraud	Med	High	Verification gates; dispute flow; human review queue for institution-verified tier
R05	Vector search quality	Low	Medium	Hybrid dense+BM25; chunk size tuning; user feedback loop in Phase 5
R06	GDPR compliance	Low	High	Footprint page + delete flow in MVP; data residency config in Phase 5
R07	Agent misuse	Med	High	Policy limits per agent; moderator agent role; abuse detection worker
R08	Port mismatch CI	Low	Low	Known gotcha: docker-compose port 5432 local, 5433 CI. Documented in STACK_VERSIONS.md
R09	Breaking dep changes	Med	Medium	STACK_VERSIONS.md as canonical reference; breaking changes logged before any upgrade
R10	Python/TS boundary	Low	Low	Queue-only communication via BullMQ/Redis; no direct cross-language calls; independent deployment

5. Success Metrics by Phase

Measurable definitions of success

Phase 1

20+ claimed entities; 0 auth bypass issues; all IDOR tests green; feed renders under 200ms p95

Phase 2

File Clerk processes PDF/HTML/DOCX; semantic retrieval returns relevant chunks; zero data leakage between entities

Phase 3

Graph traversal returns correct neighbours; graph explorer renders 100+ node graphs; edge CRUD IDOR tests green

Phase 4 — MVP

End-to-end agent session working; context profiles populated; transparency layer live; 5+ beta B2B entities onboarded

Phase 5

Token billing live; first paid enterprise contract; token margin > hosting cost per entity

Phase 6

Developer API in beta; 3+ third-party integrations; LoRA fine-tuning producing measurable quality improvement