Membria.AI · Investor Deck

AI agents that learn.

Membria is the memory & experience layer for AI agents — one causal-memory core powering products for software teams, industrial engineering and audit.

02

Company

Who We Are

Actiquest Labs

Applied AI Systems Company

Practical automation, intelligent workflows and enterprise-grade AI — built for real results.

Membria.AI

AI Operating Environment

Our core product — agents, memory, knowledge and verification combined into one enterprise-ready AI platform.

Vertical Expertise

Code · Fintech · Government · Audit · EPC

We specialize in complex domains where AI needs precision, context and process understanding.

How We Work

Applied AI Engineering

We ship enterprise AI projects that stress-test and fund Membria's core — concept to production.

03

Company Profile

Actiquest Labs at a Glance

Key facts, production footprint and enterprise AI capabilities.

10+

years enterprise product, AI engineering and venture-building

4

verticals — Fintech · Government · Audit · Engineering/EPC

100+

seats deployed across production AI use cases

Metric	Value
Founded	February 2023 · California, USA
Core Product	Membria.AI — memory & experience layer for AI agents
Production Verticals	SDLC · Fintech/Banking · AI Accounting/Audit · Engineering/EPC
Deployment	Cloud SaaS · self-hosted · on-premise enterprise
Clients & Partners	OES · Wowcube · Hiscox · K-Telco · Kazakhstan Government · industrial enterprises across EMEA/CIS
Technology Partners	NVIDIA · Amazon Web Services
Team Background	20+ years enterprise software & audit · 9 years applied AI — finance, government, industrial, deep-tech

04

Problem

Intelligence that resets every morning

1.

AI agents are stateless

Every session starts from zero. Yesterday's reasoning, decisions and failures are discarded the moment the window closes.

2.

Mistakes repeat — and they're expensive

Coding teams re-solve the same problems; in engineering, 28% of EPC budgets are lost to rework and 9 of 10 large projects overrun by +28% on average.

3.

Handovers bleed context

30–50% of assumptions vanish at every phase handover; 18% of working time is spent searching for data that already exists.

4.

Judgment walks out the door

A senior engineer leaves — and fifteen years of decisions, lessons and "don't do this" leave with them.

Every AI tool today can answer questions. None of them remembers why decisions were made — or what happened next.

05

Why now

The experience graph is unclaimed

1.

Agents became the workforce

Claude Code, Codex and Cursor now do real daily work in code, engineering and finance — but they shipped without memory.

2.

MCP became the standard port

One protocol plugs a memory plane into every agent on the market — distribution without integration projects.

3.

Regulation demands traceability

EU AI Act and ISO 19650 require defensible decisions and immutable audit trails in safety-critical work — exactly what a causal memory produces by default.

4.

Incumbents own data, not experience

Nobody owns the decision → outcome graph. Whoever captures it first gets a moat that compounds with every user, every day.

06

Product

Membria Reasoning Graph Workforce

Institutional knowledge Memory pillars

Memory

Decisions — every choice captured with its reasoning
Consistency — behavior chains keep teams aligned
Skills — wins distilled into reusable know-how
NegativeKnowledge — failures become rules

Multi-agent harness Reasoning pillars

Reasoning

Mesh — many agents work in parallel
Coordination — agents hand off via A2A
Routing — each task to the right model
Quality gates — verification blocks bad output

Domain-agnostic foundation Graph pillars

Graph

GraphRAG — semantic search over a property graph
Multi-domain — one schema across every domain
Orchestration — work runs as a dependency graph
Adaptive — memory that strengthens with use

07

How it works

One loop, every stage

Not a memory bolted onto one step — every shipped outcome feeds the next decision.

01

Plan

Recalls decisions and outcomes from similar work — the agent starts from what already worked.

02

Execute

Injects validated skills and blocks known anti-patterns as the work is being done.

03

Review

Checks the result against past failures and standards — deterministic critics, auditable output.

04

Ship

Records the real outcome and links it back to the decision that caused it.

↺ Every outcome feeds the next decision. One MCP connection: npx @membria/cli init

08

Focus

One core, many domains

Verticals are distribution of one product — not divergence. The same graph schema ships verbatim in every codebase.

Product	Who buys	Delivery	Status
Membria CE	solo developers on Claude Code / Codex / Cursor	cloud SaaS · zero-install MCP	SaaS · code.membria.ai
Membria Enterprise	engineering & product teams	self-hosted platform · own UI	Demo · design partners
AI-EPC	EPC contractors & design institutes	vertical app on the same core	Live demo · epc.membria.ai
Glass	accounting & audit teams	vertical app on the same core	Design-partner stage

Proof of substrate: Decision · Outcome · NegativeKnowledge · Skill — one schema, one causal experience loop, shared verbatim across every codebase. A vertical competitor can't follow us across domains; we can.

09

Case Studies

Proven Across 4 Domains

One AI substrate · any vertical domain

Software Development / AI-PDLC

Senior engineers leave and take their judgment. Membria keeps every decision — teams stop re-solving the same problems.

Banking / Finance

A new regulation used to mean a six-week release. Now it's live on the next transaction — traceable to the clause that authorized it.

Glass — AI Accounting / Audit

Audit teams lose weeks tracing entries to source documents. Membria marks every posting with origin and trust score.

AI-EPC / Engineering

A missed P&ID clause becomes a safety incident. Five AI critics check every diagram against standards before sign-off.

Common pattern: domain knowledge → graph memory → AI workflows → human verification → production. 100+ seats deployed across these use cases.

10

Why Membria?

Core Capabilities

The full stack — from memory and reasoning to deployment and governance

AI Agent Orchestration

Multi-agent BBS mesh, WorkGraph DAG, A2A protocol coordination

Causal Memory

Decision / Outcome / NegativeKnowledge graph with time-decay and compounding

Verification Engine

Cross-document consistency checks, rule-based normative validation, quality gates

Graph Intelligence

Memgraph property graph, symbol-level CodeGraph, policy graph with Hebbian co-activation

LLM Integration

Claude / GPT factory, structured output, behavior chains, context injection

Product Engineering

Full-stack: FastAPI + React + Node.js, MCP protocol, Telegram, Web Dashboard

Domain AI Systems

Vertical-specific schemas, critics, standards compliance (ISA / API / ASME / IEC)

Security & Governance

PolicyEngine, RBAC, OIDC/SSO, prompt injection defense, secrets management

Bottom line: Generic AI forgets. Membria remembers, reasons, and governs.

11

The wedge

EPC first: nobody remembers

Description → P&ID in minutes

Plain words or an equipment CSV in; a laid-out diagram out.

Real CAD files, not pictures

DXF for AutoCAD, DEXPI for AVEVA/Hexagon, IFC4 for BIM.

Five-discipline critics

Process, Mechanical, HSE, I&C, Piping — deterministic, auditable, repeatable.

Every defect becomes memory

Finding → change request → negative knowledge, automatically.

Zero competitors with persistent memory

Autodesk, Bentley, AVEVA, Trimble, Procore — all stateless.

Regulation is the sales agent

EU AI Act + ISO 19650 make traceable decisions mandatory in safety-critical infrastructure.

Graph data model

Every P&ID, spec and decision — one connected graph

12

Moat

Three EMPIRIA graphs, day one

Cold start is the hardest problem in agent memory. We don't have one — every vertical ships with a pre-loaded PK/NK library before a single client session runs.

Vertical	PK — best practices	NK — anti-patterns / prohibitions	Source	Status
Engineering (EPC)	~1.3M	~1.2M	20,000+ standards — ISO · IEC · ASTM · ГОСТ · API · ASME · NFPA · DNV	Live · Scaling · epc.membria.ai
Code	~250K Skills	~320K AntiPatterns	96.5K mined repos + curated sources, CWE-mapped	Live · Scaling · code.membria.ai
Audit / Glass	schema live	schema live	materiality & provenance graph — seeding from client engagements	Design-partner stage

Same schema, same loop, three domains. A generic memory startup starts every client at zero. We start every client already knowing what fails.

13

Data moat

The engineering knowledge graph no one else has

20,000+

engineering standards processed — API · ASME · IEC · ISO · ГОСТ · DNV · NFPA · AWS · ASTM

2.5M+

machine-extracted rules: ~1.2M prohibitions + ~1.3M best practices

10,000+

equipment types mapped to rules across 20+ disciplines

6

jurisdictions in one graph: US · EU · UK · Norway · CIS · International

ISO

~35,000

IEC

~10,900

ASTM

~9,200

ГОСТ (CIS)

~1,800

API

~1,760

ASME

~1,200

NFPA

~760

DNV

~700

AWS

~450

+ ACI · EEMUA · WRC · HI · FM · CGA · NORSOK · Eurocode — bars on a sqrt scale for readability

Prohibitions + best practices, one graph. No incumbent has anything like it — in any jurisdiction.

14

Proof engine

Code: evidence, not opinions

Membria CE is live for solo developers — and every session strengthens the shared graph the enterprise products run on.

96.5K

production repos mined by the CodeDigger crawler

7,448

anti-patterns cross-validated with an objective removal-rate signal

~570K

Skills + AntiPatterns live today — scaling toward 1–2M

1

command to connect: npx @membria/cli init

→

Removal-rate: the signal nobody else mines

The share of pull requests that delete a pattern, computed from commit archaeology — historical evidence from real codebases, not scraped opinions.

→

PLG funnel feeds the moat

Solo developers adopt in one command; their validated experience compounds the commons graph that enterprise and vertical products query.

15

Competition

Nobody remembers

Player	What they do	Persistent memory?	Gap
Autodesk	design + construction cloud	No	AI = copilot for CAD, not knowledge management
Bentley Systems	infrastructure digital twins	No	twin ≠ memory — no decision provenance
AVEVA / Schneider	process industry digital twins	No	operational data, not engineering decisions
Procore	Helix AI automation	No	stateless — every request starts from zero
SymphonyAI	P&ID drawing recognition	No	recognition, not a memory plane
Memory startups mem0 · Zep · Letta	generic agent memory	Yes	no vertical depth, no verification, no standards graph

The real competition is inertia: Excel + "it works fine as is." Every project starts from a blank page — that is the memory problem, and it's solvable.

16

Model & roadmap

Sequencing, not spreading

CE subscriptions for solo developers · per-seat team plans · self-hosted enterprise license · EPC per-project pricing.

Now

Foundation

1 bank + 1 government, live
EPC pilots in engineering
CE & EPC SaaS launching in parallel
100+ seats in production

Next 12 months

EPC revenue

first EPC contracts
Glass audit beta
commons → 1–2M nodes
ISO 27001 / SOC 2 path

24 months

Memory-plane API

open experience layer
any agent, any vendor
cross-industry commons
platform economics

17

Let's build the layer
agents can't work without.

Team & Ask

Let's Work Together

Michael Aprossine

CEO · Co-Founder

Scaling operations, customer-focused growth

Phillipe Khomenok

COO · Co-Founder

Strategic vision, automation-driven growth

Mike Keer

CTO · Co-Founder

AI architecture — fast, reliable, built to scale

Ask	Detail
Traction	1 bank + 1 government live enterprise · multiple EPC engineering pilots · 100+ seats deployed · CE & EPC SaaS launching in parallel
Raising	$2M seed on a $15M pre-money valuation
Use of funds	EPC go-to-market & certifications · commons scale-out to 1–2M nodes · enterprise connectors · core team
Contact	hi@actiq.ai · membria.ai · code.membria.ai · epc.membria.ai · glass.membria.ai (soon)