Seif Hemdan

Seif Mohamed Hemdan HashemDhahran, Saudi Arabia

From questionto working form.

I design and build applied and agentic AI systems, and I build and develop ventures.

See the work Open the CV
16
Project groups
#521
PSA career-high
EN/AR
Arabic and English
2026–
Independent builder
01Decisions become systemsLocal prototypes

Can a decision be tracedback to its evidence?

Founder OS / Founder Desk / SEIFO runtime. A decision workspace on local language models: interfaces, schemas and workflows that link projects, evidence and decisions. Founder Desk is described as privately hosted.

Trace a decision

3Linked records: project, evidence and decision

Decision

Treat a project, its evidence and a decision as linked records, so a decision can be traced to what supports it.

My contribution

Designed the interfaces, schemas and workflows. The Python, Streamlit and SQLAlchemy builds were AI-assisted, with local checks.

Current limit

Local work as described in the project record. Not a security certification, and not a currently operating agent fleet.

02Decisions become systemsWorking prototype · Synthetic data

Does the offer answerwhat the buyer asked for?

RFQ Review. A bilingual procurement workflow. It matches buyer requirements to supplier offers in English and Arabic, flags missing terms and drafts the clarification questions.

Check the offer

Illustration · synthetic data
EN/ARRequirements, offer and drafted clarification, in both languages

Decision

Match each requirement to the offer, keep the source figures as written, and turn each missing term into a clarification question.

My contribution

Defined the workflow. Directed the AI-assisted coding and the local checks.

Current limit

A synthetic demonstration. No buyer savings and no production deployment are claimed.

03Judgment includes revisionResearch · Local experiments

Does the claim still holdagainst a fairer baseline?

OCCIR. An AI controller was audited against a fairer baseline. The negative result was preserved, and the claim was narrowed.

  1. Claim
  2. Audit
  3. Negative result
  4. Narrowed claim
Run the audit · baseline

Illustration · no measured values
1Negative result, kept on the record

Decision

Preserve the negative result and narrow the claim.

My contribution

Audited the AI controller against a fairer baseline. Explored deterministic verification, admission controls and reproducible replay.

Current limit

No verified performance breakthrough is claimed.

Judgment includes revisionHow the work is made

How a question becomesa working form.

  1. Question
  2. Decision
  3. Working form
  4. Limit

Start from a human problem. Bound it to one decision. Give that decision a form that can be inspected: a product flow, a model, a prototype or an experiment. Then state what the work does not show yet, and the next validation step where there is one.

Grow the answer

What is mine

I am not an engineer by training. I define the problem, write the requirements, direct AI coding agents and local language models through the build, check the result, and revise when the evidence shows a limit. The code is written with AI tools. The direction, the checks and the decisions are mine.

Go to the source

I take questions directly to suppliers, professors and specifications rather than rely on summaries.

Build it to know it

I produce a model, a prototype or a requirements document, so there is something real to test.

Grade my own certainty

I record what the evidence supports and, just as clearly, what it does not.

04Products begin with peopleIn development

How can a beginner practicethe conversations they have every day?

Everyday English. A personal learning product: fifteen lessons and Arabic-supported practice, built around a beginner learner's daily conversations.

Move through the day

15Lessons, built around morning, midday and evening conversations

Decision

Start from the learner's day. Build the lessons around daily conversations, and support practice in Arabic.

My contribution

Designed the 15 lessons and the Arabic-supported practice. Developed the learning flow, the content and the progress views.

Next step

First-user feedback is the next milestone.

Current limit

The lesson count describes scope, not learning efficacy.

05Products begin with peopleHosted build · Availability unverified

How do people take part in a project,and who owns what?

HOOCH / حوش. A bilingual creative hub in English and Arabic: project pages, idea submissions, private messaging and protected owner access.

Who owns what

Illustration
3Ways to take part: project pages, idea submissions and private messaging

Decision

Make taking part explicit: who takes part, who owns which scope, and how delivery is reviewed.

My contribution

Led the build of the hosted hub. Defined participation, scope ownership and delivery review.

Next step

Verify current availability and the specific working features.

Current limit

Not an operating physical venue, and no paid membership traction is claimed. Specific working features are unverified.

06Ventures prove one thing firstResearch · Commercial discovery

Where does a better materialcreate a real advantage?

MAGNEVIA. A founder-led venture connecting advanced materials with more efficient energy and industrial technologies. Its first focus is permanent magnets, magnetic alloys and the supply chains behind them, with Saudi Arabia as the first market.

You are the magnet

7Sheets in the financial model, alongside supply-chain maps and feasibility analysis

Decision

Start from the application a material must serve. Test any advantage in performance, material use, supply reliability or economics in that one setting before going wider.

My contribution

Setting the direction and connecting the research, application and commercial questions. Built the supply-chain maps and a seven-sheet financial model.

Next milestone

Define one industrial application, document its requirements, and set a credible comparison and validation plan.

Current limit

Technical questions still need specialist review or experimental validation. No laboratory result, manufacturing or sales is claimed.

The full body of work

Sixteen project groups,where each one stands.

All 16 project groups, what each one is and where it stands. Open a row for its detail and its limits.

Systems, agents and evaluation 7 of 16

Founder OS / Founder Desk / SEIFO runtimeLocal prototypesInterfaces, schemas and workflows linking projects, evidence and decisions.

AI-assisted Python, Streamlit and SQLAlchemy builds with local checks. The project record describes Ollama/Qwen3 configuration, model-file verification, parent-child lineage, checkpoints, signed decision records, memory, handoffs and replay checks. Founder Desk is described as privately hosted. Not a security certification or a currently operating agent fleet.

RFQ ReviewWorking prototypeEnglish-Arabic workflow that matches requirements to offers, flags missing terms and drafts clarifications.

Defined the workflow, and directed AI-assisted coding and local checks while preserving source figures. A synthetic demonstration. No buyer savings or production deployment is claimed.

OCCIRResearch and experimentsAudit of an AI controller against a fairer baseline. The negative result was preserved and the claim narrowed.

Explored deterministic verification, admission controls and reproducible replay. Research and local software experiments. No verified performance breakthrough is claimed.

Mission WorkbenchPrototypesMission compiler with role presets, review checkpoints and branching decisions.

Demonstration workflows for evidence, alternatives, falsification, implementation and critique, with text and JSON export. Agent-workflow prototypes.

TRACE / Cognitive Foundry / PASS2PrototypesEvidence-workspace prototypes connecting evidence to decisions.

Source inventories, claim and event ledgers, relationships, search, and import/export prototypes. TRACE included optional encrypted packages. No certified assurance is claimed.

AETHER-K2Local experimentsSerialization, validation and encryption experiments using explicit data structures and checks.

Local data-handling experiments, with the limits of the tested environment documented.

SEIFO Universal Identity ProtocolEarly protocol workA portable identity package for consistent AI-assisted output and cross-tool handoffs.

An initial package with visual, voice and behavioral rules, source authority and model adapters. A consistency brief; early protocol work.

Products and media 5 of 16

Everyday EnglishIn developmentFifteen lessons and Arabic-supported practice built around a beginner learner's daily conversations.

Designed the lessons and the practice; developed the learning flow, content and progress views. First-user feedback is the next milestone. The lesson count describes scope, not learning efficacy.

HOOCH / حوشHosted build, availability unverifiedBilingual creative hub with project pages, idea submissions, private messaging and protected owner access.

Led the build and defined participation, scope ownership and delivery review. Current availability and specific working features require verification. Not an operating physical venue, and no paid membership traction is claimed.

DigiLaser / SallaConceptBilingual shopping concept with six selection options and routing logic.

Includes a backend handoff for later merchant integration. No completed merchant integration is claimed.

HIXEL-MOS / MediaWorkflows and prototypesNarrative, routing and approval workflows alongside film outputs and interactive visual prototypes.

Described from the project record. No footage is shown here, and no finished outcomes are claimed.

GOOD STRANGE / THREADCommercial packagingCreative service offers, a 36-month financial model, and a brief-to-content workflow with previews and reversible edits.

Commercial packaging and a production and review process. The model's duration is not a revenue result.

Ventures and research 4 of 16

MAGNEVIAResearch and commercial discoveryAdvanced materials and energy technology, starting with permanent magnets and magnetic alloys for Saudi industry.

Technical and commercial research, supply-chain maps, application requirements and feasibility analysis, with a seven-sheet financial model. Next: one industrial application, its requirements and a validation plan. No laboratory result, manufacturing or sales is claimed.

ADS / ChargeSkinBrief and factory RFQsAdhesive emergency-charger brief and factory RFQs.

Manufacturer feedback clarified drawings, assembly responsibility, thickness and quotation requirements. Manufacturability remains an engineering-review question. Safety and lab validation are not established. Rendered concepts are not physical prototypes.

Energy & materials research / ENTROPY ECONOMY / THERMAL ASSURANCEDesk and computational researchIndustrial energy loss, energy-model comparisons and cooling-assessment scopes.

Defined measurement and verification needs, separated buyer use cases, and used desk and computational work to frame the next validation step. No measured field savings, customer engagement or partnership is inferred.

OOPSConceptEssentials retail for items forgotten at everyday venues.

An essentials-retail proposition, a five-page concept, a QR and display route, and outreach material. Concept work.

Experience

Self-funded. Self-coached.Career-high #521.

I design and build working prototypes with AI coding agents and local language models, and I own each project from the problem statement to the check of the result. Before that, I played the PSA World Tour.

  1. 2026 –
    Independent Builder, Applied AI and ProductSelf-employed, Dhahran
  2. 2026
    Business Development InternKFUPM Business School · Jan – May
  3. 2023–25
    Professional Squash Player · PSA World TourCareer-high world #521 · university captain · 2023 university championship gold
  4. 2023
    BSc FinanceKing Fahd University of Petroleum & Minerals · June
  5. 2022
    Financial Analyst InternA.A. Turki Group (ATCO)
  6. 2019
    Research AssistantKFUPM · three funded research projects

Tools Claude, ChatGPT, Gemini, Ollama, Qwen3 · Python, Streamlit, SQLAlchemy, HTML/CSS, PowerShell, Excel

نبذة بالعربية

أنا سيف محمد حمدان هاشم، أقيم في الظهران بالمملكة العربية السعودية. أصمّم أنظمة الذكاء الاصطناعي وأبنيها.

حين يكون السؤال مهمًّا أحوّله إلى شيء يمكن اختباره: نموذج أولي، أو نموذج مالي، أو تجربة. أبني بتوجيه أدوات البرمجة بالذكاء الاصطناعي، وأتحقق من النتيجة، وأذكر حدودها بوضوح.

أعمل بالعربية والإنجليزية. للتواصل: saifhamdan142@gmail.com

Contact

Have a role ora project to discuss?

Reach me the way that suits you. Scan the code to save my contact card. I'm open to applied AI and product roles, venture work and research collaborations.

Save contactScan with a phone camera to add my name, numbers and email.

CV

Seif Mohamed Hemdan Hashem

Dhahran · saifhamdan142@gmail.com · +966 57 072 7440 · WhatsApp +966 55 853 0080 · linkedin.com/in/seifhemdan · seifhemdan.com

Profile

I design and build applied and agentic AI systems, and I build and develop ventures. I define the problem, write the requirements, direct AI coding agents and local language models through the build, check the result, and revise when the evidence shows a limit. I record what the evidence supports and, just as clearly, what it does not. Open to applied AI and product roles, venture work and research collaborations.

Experience

Independent Builder, Applied AI and Product · Self-employed, Dhahran · 2026 – present
I design and build working prototypes with AI coding agents and local language models, and own each project from the problem statement to the check of the result.

Business Development Intern · KFUPM Business School · Jan – May 2026
Market and competitor research, proposals and partnership investigation.

Professional Squash Player · PSA World Tour · 2023 – 2025
Self-funded and self-coached. Career-high world ranking of 521. University captain, and 2023 university championship gold.

Financial Analyst Intern · A.A. Turki Group (ATCO) · 2022
Variance analysis, budgeting, forecasting and investment-analysis support.

Research Assistant · KFUPM · 2019
Data collection, cleaning and validation across three funded research projects.

Selected work

Founder OS · Founder Desk · SEIFO runtime
Decision workspace on local language models. Local prototypes.

RFQ Review
Bilingual procurement workflow. Working prototype on synthetic data.

OCCIR
AI controller audited against a fairer baseline; negative result preserved, claim narrowed.

Everyday English
Fifteen lessons and Arabic-supported practice. In development.

HOOCH / حوش
Bilingual creative hub. Hosted build, availability unverified.

MAGNEVIA
Advanced materials venture: permanent magnets and magnetic alloys for Saudi industry. Research and commercial discovery.

Education and tools

BSc Finance · King Fahd University of Petroleum & Minerals (KFUPM) · June 2023

AI tools Claude, ChatGPT, Gemini, Ollama, Qwen3
Build and analysis Python, Streamlit, SQLAlchemy, HTML/CSS, PowerShell, Excel
Languages Arabic and English