# Mika Labs — Full Site Content > Source of truth for AI fetchers. An advanced AI research lab building the learning systems universities trust to teach. > Canonical URL: https://mikalabs.org --- ## Home — https://mikalabs.org/ MikaLabs — "Advanced AI research for education." An advanced research lab building the AI that universities trust to teach. We design, train and prove learning systems — then ship them as products. Headline pilot result: **+8.7pp** pass-rate lift across two institutional pilots. The lab splits into three frontiers: - **Mika** (flagship platform) — The AI-native learning ecosystem universities run instead of an LMS. An adaptive AI ecosystem for STEM. (mika.mikalabs.org) - **Scripply** (consumer product) — AI revision that gets GCSE & A-level students through exam season. Spec-aligned, mark-scheme aware. (scripply.ai) - **Research** (the lab) — Our research, our benchmark, our open models, and the people building MikaLabs. Publications · RedPen · Vector 1 · Team. Contact: ahmed@mikalabs.org. --- ## Mika — https://mikalabs.org/mika **Institutional STEM AI. Better STEM outcomes, at the scale of an institution.** Mika is the AI-native learning ecosystem universities and publishers use to deliver personalized STEM learning across whole curricula — built to replace outdated LMS systems, not bolt onto them. Proven across two semesters and 600+ students. At-a-glance: 600+ pilot students · 2 semesters live · subject-native STEM models. ### Proven in real classrooms 600+ students in pilot · 2 pilots completed · 60K+ platform visits · 22K+ hours of study. ### Pilot results - **+8.7pp** pass-rate lift vs. prior-semester baseline (Pilot 2, 600+ students). - **+4.35pp** average-score lift, sustained as the cohort doubled in size. - **97%** of high-achieving students are active Mika users (both pilots). ### Two ways to deploy - **For universities & schools** — the AI learning ecosystem replacing the LMS for STEM: per-institution deployment (not per-seat consumer accounts); SSO, SIS sync, and curriculum alignment from day one; faculty oversight, analytics, and gap detection. - **For content publishers** — the AI layer for STEM content: embedded deployment inside your existing product; API access to the Cartographer engine and its subject-native models; co-developed integrations with content partners. ### The Mika ecosystem - **Ask Mika** (flagship) — your AI-powered study companion. State-of-the-art chatbot on the Cartographer engine: image/document support, speech-to-text, Deep Think mode, Live Search, and live document generation. (Deep Think Mode · Live Search · Image Support · Speech-to-Text · Doc Generation) - **Mika Vision** — see everything, understand instantly. Real-time screen-aware vision; reads handwriting, diagrams and textbook pages with no upload. (Real-Time Screen Vision · Zero Upload Needed · Handwriting Recognition · Diagram Understanding) - **Mika Live** — real-time AI voice tutoring across 72+ languages, every STEM subject. (72+ Languages · All STEM Subjects · Real-Time Voice · Adaptive Difficulty) - **Mika Imagine** — documents, decks & video from one prompt. Turns a topic or your own notes into study documents, slide decks, and narrated explainer videos — spec-aligned, fully editable, generated in one pass. (Study Documents · Presentations · Explainer Videos · One-Prompt Export) - **Adaptive Practice Engine** — practice that adapts until it sticks. Auto-generated, auto-marked practice built on constructed responses (not just multiple choice); remediates with retrieval and spaced repetition, and gates on mastery. (Constructed Responses · Auto-Grading · Spaced Repetition · Mastery Gating) - **Gap & Misconception Detection** — tell a missing skill from a wrong belief. Two detectors: traces a wrong answer back to either a missing prerequisite (a gap) or a confident, coherent wrong belief (a misconception), and routes each to the right remediation. (Gap Detection · Misconception Detection · Root-Cause Analysis · Recovery Paths) - **Mika Consultant** — your AI teaching assistant. Autonomous agent: drafts lesson plans, generates assignments, flags at-risk students, reaches out via Mika Connect. (Auto Lesson Plans · Smart Assignments · Student Outreach · At-Risk Alerts) Supporting suite: Mika Docs (document editor for learning), Mika Connect (student–teacher messaging + AI live meetings), Mika Heartbeat (AI weekly study plans), Smart Flashcards (spaced repetition), Deep Analytics (per-concept insight), Productivity Suite (to-do, grade calculator, pomodoro, smart syllabus, office-hours booking). ### The engine — subject-native models "A frontier model for every subject — trained to teach." Not one general model stretched across STEM. Each subject runs on its own frontier model, purpose-built and trained on how that discipline is actually taught — then it reasons adaptively. 1. **Subject-native frontier models** — a dedicated frontier model for each STEM subject (calculus, organic chemistry, mechanics), rather than one general model stretched thin. 2. **Trained for education** — co-trained on how each discipline is actually taught: the notation, the standard methods, and the misconceptions students reliably hit. 3. **Adaptive thinking** — scales reasoning to the question: instant recall for a definition, deliberate step-by-step work for a multi-stage proof. ### Mika vs. ChatGPT vs. Claude vs. Gemini Mika is built for STEM teaching where general assistants aren't. Mika is "full" on every capability: - STEM specialised models — a subject-native frontier model for every STEM domain. - Adaptive reasoning — scales thinking depth to the problem. - Curriculum alignment — ingests syllabi, content, and rubrics. - Live voice & visual tutoring — native, real-time, on every platform, 72+ languages. - Knowledge-gap & misconception detection — per-concept, real-time, with remediation. - Auto-graded & generated assessments — AI-built and AI-marked, open-response and full working, not just multiple choice. - Faculty oversight & policy — per-course controls, audit, gap reports. - Student-data privacy — contractually never trained on. ### What Mika creates (built-in generation) Mika generates where general assistants mostly can't: narrated explainer videos (lecture-ready); images & STEM diagrams (accurately labelled, curriculum-correct); graphs & data plots (precise functions and axes from a prompt); slide decks (lecture-ready, structured); documents & worksheets (formatted, with worked solutions). ### What students said - "I love Mika. It saved me." — Calculus 2 student, Pilot 2 - "Mika is where I study Calculus 2. Everything I study is on Mika — from revising and understanding questions to practicing them before midterms and finals." — Pilot 1 - "It is the most useful platform for studying Calculus 2." — Pilot 2 --- ## For Institutions — https://mikalabs.org/institutions **AI that integrates with your curriculum, not your IT roadmap.** Mika is the institutional AI-native learning ecosystem for STEM — one contract, one deployment, every STEM course your faculty teaches. Built to replace outdated LMS systems for STEM, not bolt onto them. What you get: - **Curriculum ingestion** — Mika ingests your syllabus, course materials, and assessment rubrics so every interaction is aligned to what you actually teach. - **Identity-stack native** — SAML/OIDC SSO and SIS sync slot into your identity stack; Mika replaces the LMS for STEM courses outright. - **Faculty oversight & analytics** — course-level dashboards, gap-detection reports, and content-filter controls, under the institution's policies. Outcomes (same pilot data): +8.7pp pass-rate lift · +4.35pp average-score lift · 97% top-student adoption. Deployment scope — start with one course, scale to the institution: - **Course Pilot** (featured entry point) — pilot Mika in a single course, live in as little as 4 weeks. One-third of standard institutional cost; additional incentives on contract signing. - **Tier 1 — Department pilot** — multiple courses, one department, single semester. - **Tier 2 — School / faculty-wide** — standardize across STEM offerings; curriculum mapping, cross-course gap detection. - **Tier 3 — Institution-wide** — university-wide license; custom SIS and identity integrations. Procurement & security: data residency options (EU / US / on-prem); no model training on student data (contractually guaranteed); procurement-ready (security questionnaires, DPAs, reference architectures); security whitepaper under NDA. --- ## For Publishers — https://mikalabs.org/publishers **The AI layer for STEM content.** You have the catalog and the institutional relationships; Mika is the AI engine that embeds into your product — content-aware, deeply integrated, and built for STEM specifically. The problem: generic chatbots don't know your curriculum, point tools don't talk to each other, and publishers carry the content authority but lose the AI-driven student layer to whoever ships fastest. What we offer: - **Embedded deployment** — Mika components run inside your existing product surface; you keep the relationship and the data. - **API access to the platform** — the Cartographer router, the subject-native STEM models, gap detection, and assessment generation, as a programmatic layer. - **Content-aware models** — co-trained against your catalog so the AI references your textbooks, worked examples and problem sets. How it fits: Your content → Mika ingestion (curriculum mapping, concept-graph extraction) → Cartographer router (routes each interaction to the right specialist model) → Your students (personalized tutoring, feedback, gap detection — under your brand). Partnership models: **Integration** (Mika components inside an existing product) and **Licensing** (license the engine across multiple products in your portfolio). --- ## Pilot Results — https://mikalabs.org/results **Two semesters. 600+ students.** Pilot 2 doubled the cohort to 600+ and reproduced the outcome lift — making it more significant, not less. Headline outcomes: +8.7pp pass-rate lift vs. prior-semester baseline · +4.35pp average-score lift (sustained as the cohort doubled) · 97% top-student adoption · 93.2% highly satisfied (Pilot 2) · 71.8 Net Promoter Score · 22K+ hours of study logged. Activity: 60K+ platform visits · 10M+ tokens of math tutoring · 40K+ questions generated. Methodology: two pilots with a doubled cohort; measured pass rate and average score against prior-semester baselines, in-platform engagement, satisfaction surveys, and top-student adoption. Numbers are institutional-pilot results; per-course breakdowns and the full methodology are available under NDA. Student quotes: "Mika helped break down every confusing detail into very simplified terms." · "Mika is extremely well made and organised. The user interface is perfect." --- ## Research — https://mikalabs.org/lab The MikaLabs research-lab hub. Four areas: - **Publications** — the learning science we build on, and what we learn shipping it. - **RedPen** — an open-source benchmark for how helpful and accurate AI is in higher education. - **Vector 1** — open-weight models, designed for education and free to build on. - **Our Team** — engineering rigor and academic credibility behind every decision. --- ## Publications — https://mikalabs.org/publications Research notes and writing from the team. Two posts (blog): 1. **Why a Grade Tells You Almost Nothing: The Science Behind Mika's Diagnostic Loop** — Asma Mughrabi & Ahmed Zraiqat · 2026-05-27 · 15 min. Why a grade tells you almost nothing about *why* a student got something wrong; the difference between a gap (missing prerequisite) and a misconception (confident wrong belief); and the diagnose → confront → retest loop built on retrieval, spacing and mastery gating. Tags: pedagogy, evidence, remediation. 2. **The Science Behind Mika: Why We Built a STEM Tutor That Draws, Withholds, and Adapts** — Ahmed Zraiqat · 2026-05-27 · 12 min. What six decades of learning research says about visual, interactive, scaffolded AI tutoring — and the design choices Mika made because of it (Mayer's multimedia principle, generation effect, withholding answers, adaptive scaffolding). Tags: pedagogy, evidence, design. Full text of each article is at its URL: /publications/the-diagnostic-loop and /publications/the-science-behind-mika. --- ## RedPen Benchmark — https://mikalabs.org/redpen **An open benchmark for AI in higher education** (open-source evaluation). A benchmark for how helpful and accurate AI is in higher education — graded the way an educator grades a student, across real university-level STEM. Run it yourself, read every rubric, reproduce the numbers. Repo: https://github.com/AhmedMika/RedPen Five grading axes: **Correctness** (right at degree level, factually and methodologically) · **Targetedness** (answers the actual question and the student's specific gap) · **Level / Tone** (pitched right for a higher-ed student) · **Actionability** (a concrete next step) · **Guidance** (leads toward understanding, doesn't just hand over the answer). Results by subject (educator-graded, downloadable PDF reports): Mathematics (Calculus · Linear algebra · Real analysis) · Chemistry (Organic · Physical · Kinetics) · Physics (Mechanics · Electromagnetism · Thermodynamics) · Biology (Cell biology · Genetics · Biochemistry). More subjects coming — the benchmark is open, so anyone can add and submit one. --- ## Vector 1 — https://mikalabs.org/vector1 **Open-weight models, built to assist teachers** (not students). Vector is MikaLabs' family of open-weight models post-trained to produce the materials educators make over and over: differentiated worksheets, mark schemes, misconception guides, lesson plans, and pitched explanations across maths and the sciences. Apache 2.0, and small enough to run locally and offline. Weights/model cards on Hugging Face (https://huggingface.co/MikaLabs). At a glance: Developer — MikaLabs · License — Apache 2.0 · Language — English · Subjects — Maths · Biology · Chemistry · Physics · Runs locally & offline on modest hardware. The family (rolling out one model at a time): - **Vector-L1-4B** (4B · language · Apache 2.0) — released. "Light, version 1": the first and smallest Vector model, with quantized GGUF builds. Weights + GGUF + technical report available. - **Vector-1-14B** (14B · language) — in training. The flagship, scaling the teaching-assistant recipe to 14B. - **Vector-R1-14B** (14B · reasoning) — coming soon. Reasoning-tuned for multi-step problem solving. - **Vector-V1-8B** (8B · vision) — coming soon. Reads diagrams, handwriting and worked solutions alongside text. What it does: differentiated worksheets, exam-style mark schemes (method/answer marks shown separately), misconception guides, structured lesson plans, mixed-format questions (never defaulting to multiple choice), and holds a complex multi-part brief. Vector is a specialist, not a general chatbot — not for coding, open-ended chat, or subjects beyond maths and the sciences; high-stakes assessment should always have a teacher reviewing the output. --- ## Team — https://mikalabs.org/team **Engineering rigor. Academic credibility.** Mika Labs pairs deep engineering with active academic leadership — founded by a software engineer and a professor at Khalifa University; two pilots, 600+ students, and a single coherent product. - **Ahmed Zraiqat** — Founder & Project Lead. Software engineer; leads product vision, engineering direction, and the underlying AI architecture across every Mika module. - **Asma Mughrabi** — Co-Founder, Chief Academic Officer & Head of Institutional Partnerships. Professor at Khalifa University with 20+ years teaching university-level mathematics; owns pilot programs, student-outcome measurement, academic rigor, and institutional partnerships. Direct access by design: institutional partnerships route through Asma; product and engineering through Ahmed. --- ## Scripply (external) — https://scripply.ai Also from Mika Labs — AI-powered revision for GCSE and A-level STEM. Spec-aligned (AQA, Edexcel, OCR), mark-scheme aware, with a night-before mode. Lives at scripply.ai. --- ## Contact Request a demo from any Mika-world page ("See Mika running on your curriculum") — institution, publisher, integration and partnership conversations all start there. 48h typical response. Contact: ahmed@mikalabs.org. End-to-end encrypted; your data, your servers; zero data exposure.