Manik Maurya

Independent builder
Cognitive science researcher

Kanpur, India

CogniviaConfusionMapperTree GaneshaIn the community

How do we actually know what someone knows?

“That question pulled me into the psychology of learning & memory, then cognitive science, and eventually into building Cognivia.”

CurrentlyBuilding OutFigureResearch at IIT Kanpur

Work & research

Software, research, and things built in the world.

OutFigure

Learning infrastructure

“Getting the right answer does not necessarily mean learning happened.”

Building learning measurement infrastructure for learning platforms. I'm building ways to measure what a learner actually knows, what changed, and what lasts beyond the interaction.

Visit OutFigureAPI documentation
Inside the project

What I’m building

Send assessment events. OutFigure keeps the evidence record behind every learner-concept state, issues a probability for what the learner answers next without help, and audits that probability against the answer that arrives.

How the record works

Assessment events
Each answer, hint, piece of feedback and tutor turn, posted from your backend with a pseudonymous learner identifier. Corrections supersede; nothing is overwritten.
Evidence record
Every answer is either counted as independent evidence or set aside with a reason, under a named eligibility version. Assisted work is kept, and kept separate.
Learner-concept state
What a learner has answered unaided on a concept, how recent it is, and how far the evidence goes. Where it is thin, the state says so instead of reporting a score.
Outcome audit
Predictions are frozen when issued, then matched to the answers that arrive later. An answer that never arrives is an open case, not a failure.

The question behind it

“Not what they clicked. Not what they completed. Not whether they got one answer right. What actually changed in their head?”

Illustrative assessment

Six answers. Five correct.
How much independent evidence?

1 of 6answered without help

Example from OutFigure: six answers, five correct, five assisted. One unaided answer was correct. One observation is still not enough evidence. This is an illustrative record, not a study result.

Where it stands

Currently building. The same work on learning state, memory, and cognitive error patterns in Cognivia feeds into OutFigure.

Cognivia

Adaptive learning & research

Built Cognivia to understand a simple problem: what does a learner actually know, and why did they get something wrong?

It models learning state, memory, and cognitive error patterns instead of treating every wrong answer the same. I’ve been building it as both a research instrument and a usable learning system, with the same work feeding into my research on cognitive error taxonomy and retention.

Explore CogniviaStudy preregistration
What I built & what I learned

What I personally built

Solo built a preregistered research instrument and full-stack adaptive learning platform: React, Vite, TypeScript, Flask, and Supabase/PostgreSQL. More than 45,000 lines of production code.

Implements a four-type cognitive error classifier, per-topic personalized spaced-repetition scheduling, and a seven-variable “Learning Genome” computed from per-attempt session logs: forgetting rate, retrieval strength, retrieval latency, confidence gap, consolidation efficiency, fatigue susceptibility, and error-pattern dominance.

What I’m testing

Designed a preregistered three-arm randomized controlled trial with 90 learners to test whether error-aware adaptive learning improves delayed retention compared with correctness-only adaptation and standard instruction.

The feasibility pilot showed directional support for the core hypothesis. The full multi-school trial is in recruitment; this is not a completed efficacy result.

What I learned

The hardest thing I solved wasn't a bug — it was learning to distrust my own taxonomy.

I'd built CFI to sort every wrong answer into four buckets — Recall Failure, Partial Knowledge, Confabulation, Interference — and for weeks the inter-rater reliability numbers were garbage. Two people looking at the same error couldn't agree which bucket it belonged in. I kept tuning the rubric, adding edge cases, assuming the fix was more precision.

Then a friend, Aman, over tea he insisted on paying for, told me his cousin falls asleep in class because he skips breakfast — and asked why my model assumed every wrong answer was a cognition problem at all. That wasn't a rubric bug. It was a category error in the whole instrument: I was sharpening a ruler pointed at the wrong thing.

I added two fields to the intake form that have nothing to do with cognitive science — when the student last ate, how far they walked to get there. IRR barely moved. But the model got honest, which mattered more than the metric.

What I actually learned: the hardest bugs aren't in your code, they're in your assumptions — and the person who catches them is rarely the one you're paying to look.

ConfusionMapper

Open-source Python tool

A research tool for classifying cognitive errors and measuring whether raters agree.

Built ConfusionMapper, an open-source Python research tool for classifying cognitive error types in multiple-choice distractors and measuring inter-rater reliability. Originally built as the reliability layer for my Confusion Fingerprint Index research.

Source codePython packageCompanion preprint
Inside the tool

What I built

Implements the four-category CFI taxonomy, computes Cohen’s kappa, builds a full 4 × 4 confusion matrix across category pairs, and enforces a preregistration-compliant reliability gate before downstream data collection may proceed.

  • Recall Failure
  • Partial Knowledge
  • Confabulation
  • Interference

The technical challenge

Measuring agreement on the error categories before using those categories as a learning signal. The tool supports agreement diagnostics and reproducible session exports, so disagreement can be inspected rather than hidden inside an overall score.

Result

Published as an open-source Python package on PyPI, with archival releases on Zenodo and a companion preprint on EdArXiv. ConfusionMapper was also my capstone project for Stanford Code in Place.

The threshold above is a criterion for proceeding, not a reported study result.

Inside the lab

IIT Kanpur

Affect-based decision-making and saccadic eye-movement control.

I work on affect-based decision-making and saccadic eye-movement control, with a broader interest in how emotions and social cues shape human actions and choices.

My contribution

What I personally built

Built a free-run diagnostic version of the MonkeyLogic experiment task to isolate and reduce signal-noise accumulation during long recording sessions.

The research

Research in the Department of Cognitive Science at IIT Kanpur under Prof. K. M. Sharika, using saccadic eye-movement redirection to study parallel processing in social cognition.

Ongoing work

Co-authoring a research manuscript on the cognitive mechanisms underlying decision-making and parallel processing, while contributing to behavioural experiment design, data collection workflows, research methodology, analysis, and research presentations.

Manuscript in preparation.

Tree Ganesha

Materials & environmental design

A biodegradable, seed-embedded religious idol.

I developed a biodegradable, seed-embedded religious idol using natural river clay, plant-based colours, fish feed, and seeds as an alternative to conventional idols that contribute to water pollution after immersion.

From prototype to making kits

The problem & the build

Designed idols from natural clay embedded with plant seeds and fish feed, addressing river and land pollution from conventional Plaster of Paris idol disposal.

Taking it beyond the prototype

The project advanced through District, State, and National Level Exhibition & Project Competitions, with newspaper coverage at each stage. Later distributed making kits and shared the design with 80+ artisans across four cities.

Recognition

Received the ₹10,000 INSPIRE Award under INSPIRE-MANAK, a national innovation programme of the Department of Science & Technology, Government of India. National finalist, advanced category.

Beyond a right answer

Cognitive error & forgetting

Whether different kinds of mistakes reflect meaningfully different learning states.

I work independently on learning, memory, and cognitive error modelling, with a particular focus on whether different kinds of mistakes reflect meaningfully different learning states.

The Confusion Fingerprint Index

A framework for distinguishing different cognitive error types rather than treating every wrong answer as the same signal. The paper lays out a preregistered three-arm study testing whether error-aware adaptive learning improves delayed retention compared with correctness-only adaptation and standard instruction.

Beyond the Single Forgetting Curve

A typologically extended forgetting model in which different cognitive error types can follow different retention and relearning trajectories. The paper combines theory, Bayesian simulation, and a preregistered secondary analysis plan for a 90-participant study, testing whether error type at encoding predicts how memory changes over time.

What I built alongside the papers

Simulation and figure-generation code in R for the typological forgetting model, including a simulation output grid and figures for predicted curve divergence and the power surface.

Research record

Sole author of two SSRN preprints, cross-posted to Zenodo and EdArXiv. The study is registered on OSF and the AEA RCT Registry. A Stage 1 Registered Report is under review at Psychonomic Bulletin & Review.

These are preprints and a preregistered study, with the full trial in recruitment.

Simulation codePreregistered studyPapers

Thankinator

A bot with a sense of humour

“Originally built as a surprise gift for a professor who casually joked about wanting a thank-you counter.”

She got one. With 12 rank tiers and lore.

What it does

An open-source Discord bot that counts expressions of thanks, from “ty” and “thx” to “dhanyavaad” and “merci”, and keeps a server leaderboard. It includes personal statistics, server analytics, milestone celebrations, and both slash and prefix commands.

The build

Python, discord.py, and SQLite, with more than 40 detection patterns for abbreviations, misspellings, stretched words, and multiple languages. It stores user IDs, display names, and thank-you counts rather than message content.

Source code & setup

Beyond the screen

2015–2020
Volunteer educator

1,000+students across 40+ rural government schools

The classrooms it started in

I’ve spent several years teaching mathematics and science across rural government schools in Kanpur Dehat, working with more than 1,000 students across 40+ schools.

What started as simple teaching gradually became the place where I began noticing how differently students misunderstand the same idea. A lot of my later interest in learning, memory, cognitive errors, Cognivia, and eventually OutFigure grew from those classrooms.

Taught science, mathematics, and English using spaced repetition and active recall methods. Mentored 100+ students toward Jawahar Navodaya Vidyalaya admission and taught basic literacy to 80+ senior citizens, all unpaid.

Read my essay in Kitaab

July–August 2026
Community organizer

Safe Water Initiative

Organized a community response to groundwater contamination in Kanpur, combining public awareness with clean-water distribution. I helped explain contamination risks and safer water practices to local households, while coordinating the distribution of 2,500 units of tested clean water across 800+ households.

2024–2025
Science Club president
DPS Kalyanpur

Building a science community

As President of the Science Club, I worked across the wider school community through research, mentoring, competitions, and science outreach. I mentored students developing projects for INSPIRE-MANAK, coordinated science programming through the annual Panorama fest, presented research on secondary electron emission, and represented the school through the ERUDIT International Research Programme.

Work & results

Led an 80–100 member student science club through research and interschool competitions.

Open House

I also delivered keynote and Open House sessions to audiences of 500+ students, parents, and educators, and published articles in the school newspaper and magazine.

Spoke on how coercive schooling and large coaching systems can erode intrinsic motivation and deep learning, with an article published in the school newsletter.

Research & ERUDIT

During this period, our research work received a national first-place award, while my ERUDIT work included a 20-page SDG research paper and 19 analytical articles, under the mentorship of Dr. Peter Milne through the Beyond COP21 Symposium series.

Papers & writing

The research behind
the things I build.

Education

2021–2025

Delhi Public School,
Kalyanpur

CBSE 10+2 / Science & liberal arts

Graduated in the top 3% of my class & top 1% of my school board in the CBSE Senior Secondary Examination.

School was where a lot of my interests started colliding. I led the Science Club, worked on research and science exhibitions, spoke at Open House events, and kept drifting toward questions around learning, cognition, and how people think. Around the same time, I started building Cognivia, which began as a learning tool and gradually turned into a research project.

Class XII / 2025
96%100/100 in Painting
Class XI / 2024
95.8%100/100 in Painting
Class X / 2023
97.4%
GPA, grades 9–12
4.00 unweighted / 5.14 weightediWES verified

Currently on a gap year, working on research and building.

Honours & awards

  • INSPIRE–MANAK Award

    Department of Science & Technology, Government of India. ₹10,000 award; national finalist, advanced category, for Tree Ganesha. Progressed through district, state, and national exhibitions.

    2023–2024

  • National first place, secondary electron research

    DPS Society Science Exhibition. First place and keynote presentation.

  • INSPIRE Scholarship

    Department of Science & Technology, Government of India. Top 1% in CBSE nationwide.

    2025

  • National Talent Search (NTST)

    Rank 89.

  • National Mathematics Olympiad (NMO)

    Rank 100.

    2022

Programs &
memberships

  • Stanford Code in Place

    Student & capstone builder. Completed Stanford’s introduction to Python; final capstone project: ConfusionMapper.

    2026

  • Neuroscience Mentorship Program

    Science Mentorship Institute. Studying the fundamentals of neuroscience and conducting group research investigating how metacognition influences learning in adults.

    2026

  • Drug Discovery & Molecular Docking Workshop

    Indian Institute of Technology Kanpur. Gene editing and molecular docking.

    March 2025 / Credential EDT-IITK-CAGE-21200326002

  • American Psychological Association

    Student member and affiliate, with interests in cognitive psychology, learning, memory, and human cognition. Participating in virtual workshops, structured mentorship, and research writing.

    January 2026–present

Art & other work

Ten years of making
things by hand.

Watercolour, oil, optical illusion, 3D lettering, and putty pop art.

Fine arts practitioner for more than 10 years, including commissioned illustration work for Gridline Studios.

I also kept up a long-running interest in fine arts, which probably explains why I still care as much about how something feels and looks as how it works.

Family business

Unpaid operations and customer care experience for my father’s business, balanced alongside gap-year research and academics.

Tools & interests

Research & analysis
Python, Jamovi, statistical analysis, preregistration, OSF workflows, inter-rater reliability, eye-tracking, EEG, MonkeyLogic, MATLAB, and PsychoPy Builder.
Development
React, Vite, TypeScript, Flask, FastAPI, Supabase, PostgreSQL, SQLite, Twilio API, Git, and R.
Design & working tools
Adobe Suite, Canva, and Google Workspace.
Questions I keep returning to
Cognitive science, neuroscience, learning science, memory, metacognition, psychology, human–AI interfaces, and human decision-making.
Causes
Education, science and technology, the environment, arts and culture.

Languages

English
Native or bilingual proficiency
Hindi
Native or bilingual proficiency
Latin
Limited working proficiency
Sanskrit
Elementary proficiency

A little about me

Kanpur, India
He / him

“If a question pulls me somewhere else, I usually follow it.”

I’ve probably changed directions more times than I’ve stayed on one path. I’m not very good at continuing something just because I’m supposed to.

I like building things, questioning rules that don’t make sense, asking questions that are probably too big, meeting people who think differently, and occasionally spending unreasonable amounts of time looking at cars.

I still don’t know exactly what I want to be doing ten years from now. I’m not even completely sure what I’ll be obsessed with two years from now. I think I like it that way. I’ve always felt most comfortable being a bit of an outlier anyway.

The future feels a little like my own life right now: unpredictable, changing fast, and impossible to map too far ahead.

But I keep coming back to the same thing: what does it mean to stay deeply human while everything around us gets more capable?

I want to understand how we learn, what we are good at, what makes us different, and how we build things that make more room for the parts of being human that actually matter.

Most of life still comes down to people anyway.

I have no perfect plan.
That is probably the plan.

GitHubLinkedInORCID