1. What are you making?

I’m building a real-time reading support system for students with dyslexia that adapts to how a learner is processing text in the moment.

The goal is to create a system that can detect when a student may be experiencing cognitive overload while reading, before that struggle becomes obvious through a wrong answer, disengagement, or frustration. The system will combine EEG signals with behavioral data from a reading interface, such as pauses, rereading, and response latency.

When the system detects that a student may be struggling, an AI agent will provide support in real time. Depending on the situation, it may simplify a sentence, break content into smaller chunks, switch text to audio, or help decode a difficult word.

This project is motivated by an experience I had when I was 13, teaching a class of around 30 students with learning disabilities how to code. Many students quietly struggled with reading-heavy instructions, and often the difficulty only became visible after they had already fallen behind. This project starts from the idea that learning tools should respond earlier — not after failure happens, but while difficulty is building.

This is a research prototype, not a diagnostic or medical tool. The broader aim is to explore whether accessible EEG hardware can help create more responsive, personalized learning systems.

2. How are OpenBCI tools being applied?

OpenBCI provides the full sensing setup for this project, from EEG acquisition to real-time intervention. I will use the OpenBCI Cyton 8-channel biosensing board as the main recording device while participants complete reading tasks on a computer. The Cyton will stream multi-channel EEG data at 250 Hz to a real-time software pipeline, where the neural data will be synchronized with behavioral signals from the reading interface, such as pause duration, rereading frequency, and response latency. This makes it possible to connect changes in brain activity to specific moments of difficulty during reading.

For electrode placement, I will use the OpenBCI headband kit to create a wearable setup in a simplified 10-20 configuration. The channels will be positioned over frontal, temporal, and occipital regions to capture activity related to attention, language processing, and visual processing during reading. I will also run controlled sessions using OpenBCI gold cup electrodes with 10-20 conductive paste to achieve lower impedance and more stable recordings. Comparing the headband setup with the gold cup setup will help evaluate the tradeoff between ease of use and signal quality.

To improve robustness, I will use the EMG/ECG snap electrode cables as auxiliary channels for artifact monitoring. These channels can help identify noise sources such as facial muscle activity, movement, or other physiological interference that may contaminate the EEG signal during reading tasks. After acquisition, the EEG data will pass through a preprocessing pipeline that includes filtering, artifact attenuation, and rolling-window feature extraction.

The real-time model will focus on EEG features commonly associated with cognitive load, especially spectral power changes in the theta, alpha, and beta bands. Because EEG varies significantly across individuals, each session will begin with a short calibration phase to establish a personalized baseline. The system will then combine EEG-derived features with behavioral data from the reading interface to estimate whether a learner is engaged, overloaded, or disengaged.

When the system detects a high-probability overload state, the AI agent will trigger an adaptive intervention. Depending on the situation, it may simplify text, break a passage into smaller chunks, switch to audio, or provide phonetic support. In this way, OpenBCI is not just being used to record brain activity. It serves as the real-time sensing layer in a closed-loop system that continuously estimates cognitive state and helps decide when support should be delivered.

A major goal is to make this setup reproducible for the OpenBCI community. I plan to document the full workflow, including hardware configuration, electrode placement, signal quality checks, synchronization with the reading interface, preprocessing, feature extraction, and the intervention logic, so that other builders can replicate and extend the system. 

3. Why is this important?

Dyslexia is not a deficit in intelligence. It is a difference in how the brain processes written language, and reading can require significantly more cognitive effort for students with dyslexia.

Many educational tools respond only after a student answers incorrectly, stops engaging, or asks for help. But in many real learning environments, students may struggle silently. By the time the system notices, the learner may already be frustrated or behind.

This project explores a different approach: using passive brain-computer interfaces to detect internal cognitive strain earlier and respond in real time.

Even partial success would be valuable. It could help identify which EEG signals are robust enough to support real-world educational applications, where OpenBCI tools can be used outside traditional neuroscience labs. The project also aims to make BCI development more accessible by showing how raw EEG signals can become part of a full closed-loop system: sensing, modeling, intervention, and evaluation.

I’m especially excited about this project because it connects neuroscience, accessibility, AI agents, and education. If successful, it could help move EEG-based learning systems from research concepts toward practical tools that support students when they need help most.

4. Who is involved in this project?

My name is Krishiv Thakuria, and I’m completing this project as an independent researcher. My background includes research engineering work in the Kellis Lab at MIT CSAIL under Professor Manolis Kellis, as well as co-creating and serving as an instructor for MIT 6.S189, Foundations and Frontiers of Generative AI, a for-credit MIT EECS course on AI agents.

You can find me @KrishivThakuria on Twitter! Feel free to follow me there for project updates.