Abdolmadjid Masoomi

Projects

Systems that were built and then had to keep running. Each entry says what it is, what it taught, and what is not public about it.

No user counts, revenue figures or customer names appear here. Where something is not public, the entry says so rather than implying more than can be shown.

  • Mira

    operating

    A multi-tenant AI assistant sold to customers and reached over messaging rather than a web app.

    Why it exists
    Most assistant products assume the customer will come to a website and stay there. Mira starts from where people already are, which changes almost every architectural decision behind it.
    Role
    Founder, architect and operator. Design, implementation and the running of it in production.
    What it taught
    Holding one long-lived connection per customer inside a single process is a resident model, not a request model. It cannot be load balanced, only sharded, and that constraint has to be designed for from the beginning rather than discovered later.
    Public evidence
    The product is sold publicly. Internal architecture and customer numbers are not published.
    Multi-tenant architectureLong-lived per-customer sessionsMessage-channel integrationBilling and subscription handlingGrounding responses in retrievable state

  • Context Ledger

    internal

    Append-only project memory with provenance, used as the system of record across the work described on this site.

    Why it exists
    Conversation history kept losing the decisions that mattered. What was needed was a record that outlived the conversation and could be queried rather than remembered.
    Role
    Designer and sole implementer.
    What it taught
    Recording a decision is not the hard part. The hard part is making it costly to record a claim that has not been checked, because the path of least resistance is always to write down what you believe.
    Public evidence
    Conceptual model published on this site. Schema and implementation are not public.
    Append-only recordsProvenance and evidence attached to every entryTamper-evident storageSeparation of verified fact from assumption

  • Linear Anti-Hallucination Protocol

    in development

    A governing discipline for keeping long-running work anchored to evidence and stated intention.

    Why it exists
    Confidence and correctness are not the same signal, and over a long project the difference compounds into work built on claims nobody checked.
    Role
    Originator and implementer.
    What it taught
    A rule that can be talked around is not a control. Enforcement has to be mechanical, because the moment it depends on someone choosing to comply, it stops applying exactly when it is needed most.
    Public evidence
    Problem statement published on this site. Implementation withheld pending intellectual-property review.
    Evidence gatingIntention preservationDrift detection

  • Security tooling

    operating

    A consumer security product, sold by subscription.

    Why it exists
    A gap between what security software claims to check and what it can actually observe on a given machine.
    Role
    Founder, architect and operator.
    What it taught
    A finding a user cannot act on is noise. The reporting is as much of the product as the detection.
    Public evidence
    Sold publicly. Detection methods and customer numbers are not published.
    Desktop and mobile clientsSubscription billingScanning and reporting

  • File infrastructure

    operating

    Private file storage and sharing, self-hosted rather than resold.

    Why it exists
    Wanting storage whose trust model could be stated precisely rather than assumed.
    Role
    Founder, architect and operator.
    What it taught
    If the server can deliver the code that does the encryption, then browser-side encryption is a weaker claim than it sounds. Saying so plainly is better than shipping the stronger-sounding sentence.
    Public evidence
    Publicly reachable. Internals are not published.
    Storage and deliveryAccess controlSelf-hosted deployment

  • Commerce and automation systems

    archived

    Business systems built across trading, marketing and operations before the AI work.

    Why it exists
    Ordinary commercial need. Processes that were being done by hand hundreds of times.
    Role
    Builder and operator, alongside running the business itself.
    What it taught
    Software that is elegant but unreliable costs somebody real money. That is a different standard from software that is merely interesting, and it is the standard the later research inherited.
    Public evidence
    Not published. Described here as background rather than as a portfolio piece.
    Database-backed business systemsProcess automationSearch and digital marketing operations