Skip to content
Hussain Sharifi Independent research and special projects
MenuClose

Question / Evidence / Challenge / Decision

Research that ends in a decision you can defend.

Businesses, investors, institutions, families and private principals bring a question that matters and has no clean answer.

I find the information that exists, test it against itself, and hand back what is established, what is still uncertain, and what to do next.

Bring a question See how the work runs

Nothing here is sold from a page. You describe the problem, the scope is discussed, then the work and the price are agreed in writing.

One inquiry, four moves

  1. Question

    Two suppliers, two very different quotes, and one claim I cannot verify. Which do I choose?

  2. Evidence

    Filed accounts, both contracts, three named references and the public register. Six sources, each recorded as what it is.

  3. Challenge

    Two of the references trace back to the same person. The cheaper quote excludes the work that caused the last overrun.

  4. Decision

    One recommended option, the two conditions it depends on, and the one fact still unknown, written where you can check it.

Illustrative. A composite of the shape of the work, not a client.

01 What people bring

The problems that arrive here have a shape.

They are not requests for advice. They are situations where the facts are incomplete, the stakes are real, and nobody has yet put the whole thing in one place.

  • We are about to commit, and I cannot verify half of what we have been told.

    Business research

  • Four appointments, four explanations, and nobody has looked at all of it together.

    Private health research

  • I want a model working on my own files without any of it leaving the building.

    Private AI and research systems

  • The two reports disagree, the decision is on Friday, and I need to know which one to believe.

    Business research

  • The scan was clear and the symptoms are not. What should I be asking at the next appointment?

    Private health research

02 Three practices

One method, three kinds of question.

The practices are separate because the clients, the sources and the risks are separate. What they share is how a claim is treated before it is allowed to matter.

Business research and special projects

Something important you need to understand, find, compare or verify.

Companies, founders, investors, institutions and private principals bring matters where the facts are incomplete, the consequences are real and discretion is expected.

  • Special projects, run end to end
  • Commercial and counterparty research
  • Provider research and comparison
  • Source checking and document control
  • Cross-border coordination and practical support

This is research and coordination. It is not regulated due diligence, legal practice, audit or financial advice, and where a matter needs one of those it is said plainly and the work stops there.

See the business practice

Illustrative comparison ledger. A composite, not a client.
What was claimedWhere it came fromStatus
Ten years operating in this market The company website and the pitch deck Claimed
Turnover grew in the last filed year Accounts filed at Companies House Verified
Regulated to carry out this activity The public register, searched. No matching entry found. Unknown
The delay was the supplier's fault Two contracts, one signed after the delay had begun Inferred

The ledger is the deliverable, not a summary of it. Every row can be checked, and the row that could not be settled says so instead of being quietly dropped.

Private health research

A difficult health question, researched properly and privately.

For individuals and families facing a decision the appointment did not have time to settle: records organised, evidence mapped, contradictions surfaced, and the questions worth asking prepared in order.

  • Literature and source research on one agreed question
  • Evidence mapping and intervention comparison
  • Mechanistic analysis, stated as mechanism rather than proof
  • Organising a health history so it can be read at once
  • Preparing the questions for the next appointment

Research and decision support, not medical practice. It works alongside the people treating you and never in place of them. No diagnosis is offered and no treatment is recommended.

See the health practice

Illustrative evidence and mechanism map. A composite, not a client.

  • Low ferritin explains the exhaustion, even though the blood count is normal.

    Mechanism
    Iron stores fall before the red cell count changes, so tissue supply can be short while the count still reads normal.
    Evidence
    Established
    What would settle it
    A repeat ferritin read together with inflammatory markers, because inflammation raises ferritin and can hide a low store.

  • The gut is driving the mood symptoms.

    Mechanism
    Vagal signalling, microbial metabolites and immune messengers all run in both directions between gut and brain.
    Evidence
    Mechanistically plausible
    What would settle it
    Human trials that change the gut and then measure mood, rather than the reverse. The ones that exist are small and short.

Every line carries its own state. Nothing is promoted from plausible to established because it would be more useful if it were.

Private AI and research systems

Controlled AI around your own information, with the boundary written down.

The value is not the model. It is knowing exactly what it can see, what it can do, and what it can never touch unless a person says so.

  • Local or private setup on hardware you control
  • Controlled knowledge and research workspaces
  • Project state that records where every answer came from
  • Permission design and human review gates
  • Written documentation of the whole arrangement

This is not an autonomous agent and it is not a promise that a model will run your business. Where something would need engineering, procurement or a regulated permission, that is stated at the time rather than implied now.

Read the private AI practice

The reference system that exists and runs today. Verified 2026-08-10.

Inside the boundary

  • The model One model, running on one machine, on its own hardware. Local
  • Project information Only the material placed in the project. Nothing else is visible to it. Selected
  • Source labelling Every answer names which of those files it came from. Local
  • Modes Written instructions you can read, change or delete. Read only

Outside the boundary

  • The web and current data No browser, no search, no live or current information of any kind. No access
  • Email and messages It cannot read them and it cannot send them. No access
  • Payments No payment authority of any kind exists in the system. No access
  • Publishing and changing files Nothing leaves and nothing changes without a person doing it. Approval required

That is a description of a system that runs, not a proposal. Anything beyond this boundary is something that would have to be configured or built, and that difference is stated before any work is agreed.

03 One method

Question, evidence, challenge, decision.

The same five moves, whether the subject is a counterparty, a laboratory result or a permission boundary. It is the part of the work that does not change with the subject.

  1. Move one Frame the question

    Turn what you are worried about into something that can actually be answered, attached to the decision it feeds and the date it is needed by.

  2. Move two Build the source map

    Find what exists and record what each source is. A filed document, a trial, a contract and somebody's assertion are four different things and are never levelled into one.

  3. Move three Test the claims

    Look for what would make the answer wrong: figures that disagree, a source quoting another source, a result nobody has repeated, an incentive nobody declared.

  4. Move four Organise the evidence

    Set out what is established, what is contested and what is unknown, so the shape of the uncertainty is visible rather than smoothed away.

  5. Move five Deliver a usable next step

    A written answer with named sources, stated limits and the clearest next action. Judgement and accountability stay with a person, and that person is me.

04 Selected inquiry types

The kinds of question that come in.

Written as types rather than as cases. No results are published here, because outcomes belong to the people who paid for them.

Business research

Is this counterparty what it says it is, before we commit?

Sources used
Filed accounts, public registers, enforcement and court listings, the counterparty's own documents, and named references taken directly.
Where the uncertainty sat
Ownership above the company that signs, and whether one permission actually covers the activity in question.
Form of the output
A source file and a comparison ledger, every row labelled verified, claimed, inferred or unknown.

Private health research

Two of the results disagree. What should be asked at the next appointment?

Sources used
The records themselves, the relevant national guidance, the primary literature behind that guidance, and the laboratories' own method notes.
Where the uncertainty sat
Whether the difference is a real change over time or two laboratories measuring in two different ways.
Form of the output
A written brief with the evidence state on every line, and the questions to ask, in the order they are worth asking.

Private AI and research systems

Can we use AI on this material without it leaving our control?

Sources used
The material itself, the hardware already available, the existing way of working, and the obligations that constrain all three.
Where the uncertainty sat
Which steps genuinely need a person's approval, and which are being reviewed out of habit rather than out of risk.
Form of the output
A written permission map, a description of what a controlled setup would and would not do, and a plain list of what has not been built.

05 Proof of method

What you actually receive, and how a source is labelled.

Two things are worth showing before anyone writes: the shape of a written answer, and the vocabulary used to say how much a claim is worth.

The shape of a written answer

  1. The question, written as it was agreed, so it cannot drift.
  2. What is established, with the source beside each line.
  3. What is contested, and who says what.
  4. What is unknown, and what it would take to know it.
  5. The realistic options, and what each one costs you in risk rather than in money.
  6. The clearest next step, and the date after which it stops being the right one.

How a source is labelled

  • VerifiedChecked against a primary record, and the record is named.
  • ClaimedSomeone asserts it. Recorded as an assertion, with who is asserting it.
  • InferredReasoned from what is known, with the reasoning shown so you can disagree with it.
  • UnknownLooked for, not found, and said so rather than left out.

AI accelerates searching, reading and drafting. It does not decide what a source is worth and it does not sign the answer. One person does both, and that is the part that is never automated.

06 Research

Published research, written and reviewed here.

The library is considerably larger than this. These five entries are the ones that best show how evidence is handled when the honest answer is not the popular one.

Browse the full research library

07 About and accountability

One person does the work and answers for it.

I am Hussain Sharifi. I take the work on personally and I answer for it personally. There is no team, no account manager and no queue.

First-Class Honours in Politics and International Relations from SOAS, University of London. The work since has been special projects, research and operations, some of it confidential, across more than one country. Based in London, and the work can be cross-border.

Technology can accelerate enquiry. It cannot supply judgement. Work is undertaken where claims require scrutiny, consequences must be understood and discretion matters. Evidence informs the conclusion. Uncertainty remains explicit. Responsibility remains human.

08 Bring a question

What are you trying to understand, decide or get done?

Describe it in your own words. There is no form to complete first, nothing is booked by writing, and no price is discussed before the problem is understood.

  • Business research and special projects

    State what is happening, what has been tried, the decision or deadline it feeds, and where the facts are still unclear.

    Send the issue or task

  • Private health research

    Start with the question itself rather than an introduction. Please do not send records, results or images by email; if work goes ahead, a sharing route is agreed in writing first.

    Open the private health desk

  • Private AI and research systems

    Describe the material, the hardware you have and the constraint that matters most. This practice does not have its own page yet, so it is answered from the same desk.

    Write about a systems question

Scope is discussed before anything is agreed, and the work and the price are then agreed in writing. Sending an enquiry does not create an engagement.