What we’re building now, and the track record behind it.
Our products
Built and owned by Mostech Ventures.
AI debate platform
Our product, in pre-launch
A real-time platform where people and AI agents debate, and an AI or human moderator judges every argument. Its value is judgment anyone can check: every score comes with a written rationale, named fallacies and sources.
Challenge
Let people and AI agents debate, and judge every argument in a way anyone can check.
Solution
A real-time debate platform where humans and AI agents debate, and an AI or human moderator judges every argument
Multi-agent orchestration with configurable AI personas, written by a meta-prompt agent
Schema-enforced, explainable judgments: a 0–100 score, a written rationale, named logical fallacies, inaccurate claims and sources
Trust
AI decisions drive business logic under fixed rules and an audit trail, with a human in the loop by design.
Next
A debate coach, debate practice, and a courtroom practice tool with judge, jury, prosecution, defence and defendant.
An AI moderator’s judgment of one argument: a score out of 100, a written rationale, named logical fallacies, an inaccurate claim and sources
A live, gamified debate between a person and an AI persona, with health bars, a turn timer and each argument scored
A free-form debate with three people, an AI participant and an AI moderator
Public debates, with search and live, starting and concluded filters
Requests to join a debate as participant or moderator, approved or declined from the notifications
Demo data.
Coaching platform
Our product: an AI life coach, in English and Persian
An AI life coach, in English and Persian, that follows a proven coaching method step by step. A state machine runs the method and the AI holds the conversation, so coaching stays structured and safe.
Challenge
Coaching that follows a proven, structured method (Choice Theory and Reality Therapy, WDEP), not a free-running chatbot.
Solution
“Strict machine, loose conversation”: a deterministic state machine controls every step, and the LLM holds the conversation
Scored gates that no prompt or model output can bypass
Four coach personas and validated psychometric tests
Trust
Personal details are tokenized before they reach the model. A full audit trail, and experts edit prompts without a deploy.
A live coaching session: the method’s 18 stages across the top, and a banner showing six identifiers redacted before reaching the model
The privacy audit trace: each token mapped to its original, the anonymised prompt actually sent, and the check that no original value left
A scored gate: the session moves on only when the person rates the goal at the required level
Choosing a coach: four personas, one method
The coachee dashboard: choice-to-action rate, the five basic needs, and commitments kept and missed
Demo data, fictional people.
Current work
Platforms we are building for clients.
Operations and traceability platform
For a multi-location hospitality business in Ontario
An inventory and costing platform for a multi-location hospitality business. It shows the true cost of everything produced at every location, and traces any item from supplier to customer with QR labels.
Challenge
Know the true cost of everything produced at every location, and trace any item from supplier to customer.
Solution
Inventory and production costing across purchase, raw materials, production, finished goods and sale
Moving weighted-average costing, multi-level versioned recipes, and price and quantity variances
Maker/checker approvals; bilingual in English and Persian, right to left included
Traceability
QR lot traceability from supplier to customer: labels, scanning, complaint tracing and recalls.
Next
Designed to become a multi-tenant SaaS for other businesses in the sector.
A production lot: QR code and labels, where it went for a recall, scans at the line, and the lots it was made from
Maker/checker: a checker approves or rejects each line of an entry, with a reason
One raw material’s ledger: every purchase and issue, with the moving weighted-average cost after each
The month so far: batches, total variance, stock value, losses in transfers, and variance by product
The same inventory in Persian, right to left
Demo data.
Clinical psychology platform
For a psychology clinic
A modern rebuild of a psychology clinic’s online system. It brings psychometric tests, booking, video sessions and payments into one platform, with AI transcripts of each session.
Challenge
A modern rebuild of an online clinic system.
Solution
Psychometric tests with scoring and reports, and diagnostic scenarios
Booking, video sessions and payments
AI session transcripts and topic creation
Track record
Platforms our founder led before founding Mostech Ventures.
MEERA CENTRUM (Meera Simulation)
Energy, 2019–2022. Our founder was lead architect and lead developer
AI-powered reservoir simulation for the oil and gas industry. Coupling a physics solver with a deep neural network turned months of engineering studies into days, so field-development decisions came far sooner.
Challenge
Reservoir engineering studies and field-development decisions that took months.
Solution
An industrial PDE solver coupled with a deep neural network: the first professional, industrial AI-powered reservoir simulation solution. It cut engineering timeframes from months to weeks to days.
Impact
A field-development decision cut from 18 months to under 2
Simulation runs 6–10× faster, with 33–60% less engineering effort per study
More than US$150M in operating costs saved on one asset
About US$220M of reserve over-estimate corrected, and more than US$1.6B in added revenue across its users
Source
Figures as published by MEERA in its case studies.
A 3D reservoir model coloured by porosity, with the wells
How far the AI forecast was from the simulation, per well and per month
The deep neural network: an LSTM layer, dropout and fully connected layers
Training the network: loss per epoch, and the run finishing in under nine seconds of CPU time
Correlation analysis: which variables drive water production, and how strongly at each well
A synthetic tutorial model. Coordinates removed.
Tawteen
A government. Our founder was lead architect, lead developer and lead AI architect
A government’s national labour-market system. It brings every work permit in the country into one platform, so each ministry and company sees its own workforce, with AI insights and succession plans.
Challenge
Manage a national labour market from one system.
Solution
A national labour-market management system covering all work permits in the country
Role-based views: each ministry sees its own sectors, each company its own data
AI-driven workforce insights at national and company level, and AI-generated plans and succession strategies
Impact
More than 800,000 work permits ingested into one national system, in production six months after the project started.
National map: every company plotted and coloured by its workforce alignment score. Figures removed
National view: companies grouped into 13 health-score bands, and the occupations and activities where expatriate pay dominates. Figures removed
Drill-down from a sector to its top 20 occupations by succession opportunities. Figures removed
Succession opportunities in every sector, and the top ten countries of origin of the workforce. Figures removed
Workforce growth and turnover over time, nationals against expatriates. Figures removed
Succession opportunities by sector and governorate, as a heatmap. Figures blurred
Hospital Information System, a national Ministry of Health. Our founder architected, built and delivered the AI end to end
AI services in the Hospital Information System of a national Ministry of Health. They help doctors read medical images and check prescriptions against the patient’s history, with the AI measured against expert judgment.
Challenge
Bring AI into diagnosis and prescribing safely, and measure it.
Solution
Multimodal medical-image diagnosis (X-ray, CT, MRI, ultrasound and mammography) with GPT-4o vision
Prescription-safety analysis against patient history, with unknown drugs enriched from OpenFDA and RxNorm
An independent LLM judge that scores AI diagnoses against expert-labelled cases
Impact
Shipped in the national system, not a pilot: AI diagnoses measured against expert judgment, and every AI call traced with its cost.
Petroleum portfolio management
Energy, 2019–2022
A management platform for a global and national petroleum portfolio. It unified asset data into always-current dashboards and analytics, so the whole portfolio could be managed from one view.
Challenge
Manage a global and national petroleum portfolio from one place.
Solution
Unified asset data
“Evergreen” real-time dashboards
Portfolio analytics
Impact
A single, always-current view of the portfolio.
Investment ranking: each reservoir placed by unit technical cost against value, and by value per investment against production
Actual production against the forecasts made in earlier months, the “evergreen” view
Production history with P10, P50 and P90 forecasts, next to targets and actuals
Production by equity share: by reservoir, by field and by well
Change in proven oil reserves: from the opening balance through revisions, improved recovery, discoveries, reclassifications and yearly production
Asset names are generic.
RETINA
Enterprise engineering platform
An enterprise platform for reservoir simulation and engineering data. It matched the industry’s leading tools keyword for keyword and read all their formats, and its technology demonstration attracted international investment.
Challenge
An engineering platform that stands up to the industry-leading reservoir tools.
Solution
Simulation and data platforms built to parity with the industry-leading reservoir simulation and modelling tools: about 650 simulator keywords, full import and export
Ingestion of all major industry formats, including binary and proprietary ones
Perforation simulation benchmarked against industry-leading proprietary models
Impact
Its technology demonstration attracted international investment.
A 3D corner-point reservoir grid coloured by permeability, with the wells
Oil production rate for every well over eight years of history
A fluid (PVT) model: the table and oil viscosity against pressure
Relative permeability and capillary pressure, as a table and a chart
Perforation design: a 3D view of wellbore, cement and formation, with the perforation tunnels
Field data: Equinor’s Volve open dataset. Well names and coordinates removed.
GPU reservoir simulator
National petroleum research institute, 2012
A 2D reservoir simulator optimized for graphics processors, built at a national petroleum research institute in 2012. It moved the numerical work onto the GPU to speed up reservoir simulation.
Challenge
Use GPU computing to speed up reservoir simulation.
Solution
A GPU-optimized 2D reservoir simulator, as lead developer
