SATSam

• 

  SATSam's homepage.

• 

  SATSam's Error Fingerprint feature.

• 

  SATSam's insights tab.

• 

  SATSam's personalized study schedule.

Inspiration

During junior year of high school, many students, our team included, had to take on the arduous struggle of studying for the SAT. Obviously, paying for high quality tutors that charged $200+ per hour was out of the question, so we scoured the internet for free resources. Although we found tools such as Bluebook and the SAT Question Bank to be incredibly helpful, these resources were often quite fragmented, inconvenient, and complicated to use. For example, Bluebook sometimes reused its questions in its practice tests, and the Question Bank was quite cluttered and required a great deal of time to access the right questions. We wanted to create something more personal, encouraging, and accessible. So, in this hackathon, when we noticed that the topic was going to be artificial intelligence intertwining with education, we thought that an immersive, technically advanced SAT agent was the perfect choice.

What it does

SATSam unites the essential parts of SAT preparation into one cohesive platform.

Students can create adaptive practice sessions by choosing a subject, skill focus, starting difficulty, and number of questions; or, they can allow SATSam to target their highest impact weaknesses automatically. After every response, SATSam updates the difficulty; it does not only depend on whether the student gets the question wrong or not, but their previous error patterns, proficiency in that skill, and the time spent.

Next, instead of only displaying whether an answer was correct, SATSam uses a local AI tutor named Sam to provide a detailed, personalized explanation on why the answer was wrong as well as the student's possible reasoning error. When a student misses a question, Sam generates an Error Fingerprint that analyzes the likely reason behind the mistake: this includes patterns such as a concept gap, misread question, rushed answer, tempting distractor, and much more. To be more exact, each fingerprint includes:

The likely root cause of the mistake Evidence from the answer choice The exact micro skill that needs improvement Practical strategies for future related questions Comparison with previous mistakes Additionally, SATSam also includes an Insights dashboard containing estimated Math and Reading and Writing scores, average pacing, score trajectory, strengths, weaknesses, and skill specific feedback.

Moving on, SATSam can also generate a detailed weekly study plan: each day is divided into measurable blocks such as micro lessons, targeted drills, confidence practice, and error review.

Lastly, students can open Chat with Sam from any page or directly from a missed question in order to receive guidance, feedback, or explanation on a certain question or topic. When opened from practice, the conversation is automatically grounded in that question and the student’s selected answer in order to give accurate and insightful tutoring.

How we built it

We built SATSam with Python and Streamlit using a modular architecture. Separate modules manage adaptive question selection, performance metrics, settings, AI prompts, explanations, Error Fingerprints, study plan generation, and more.

Practice questions are loaded from a structured JSON question bank containing each question's domain, skill, difficulty, answer choices, and explanation. During a session, the AI records every response, including accuracy, selected answer, difficulty, time spent, and generated Error Fingerprint. These records power numerous insights such as skill diagnostics, score estimates, adaptive recommendations, future study plans, and much more.

For the AI layer, we integrated the language model Qwen from Ollama to run directly on the student's computer. We designed separate, section specific prompts for answer explanations, error diagnosis, session reviews, domain recommendations, and weekly planning.

Lastly, we used custom HTML and CSS to create SATSam's terracotta and sage interface in order to elicit a cozy and encouraging study vibe.

Challenges we ran into

Our biggest challenge by far was altering the AI so that it contained meaningful insights for the user instead of it being a generic chatbot. Early responses were vague, overly confident, or unrelated to the student's mistake. We managed to solve this issue by grounding prompts in the question involved, referencing the explanation, and acknowledging the user's recent mistake history.

Another challenge was generating reasonable study plans. Originally, they contained missing days, invalid JSON, unreasonable question counts, and did not match the user's allotted time. We were able to resolve this issue by creating a validation pipeline that checks every generated field and reconstructs invalid portions while simultaneously preserving the AI's overall instructional intent.

Accomplishments that we're proud of

We are most proud that SATSam is not simply an AI chat interface, but a complete learning loop as it has practice, diagnosis, explanation, planning, and improvement in order.

We are also very proud of the AI Error Fingerprint feature because it gives detailed, question specific insight on the why a student got that certain question wrong. Instead of only conforming to the standard SAT preparation method of only giving the explanation for the right answer, SATSam also gives the conceptual reasoning behind why a student made that error.

What we learned

We learned that successful educational AI requires far more than merely being a chatbot. The best system combines AI judgement with structured student data, deterministic algorithms, and thoughtful design.

Also, we realized the importance of personalization and emotional design in education. Students are much more likely to continue practicing when they feel like their mistakes are being acknowledged, not criticized. Additionally, these mistakes must not be given a generic explanation that fits everyone, but a user specific analysis that encourages the student without hiding areas that need improvement.

What's next for SATSam

Our next step is to expand SATSam with a larger question bank, improved score calibration, and more user performance analytics.

We also plan to add student accounts, teacher dashboards, and mobile support. Over time, SATSam will develop an increasingly more accurate and detailed learning profile on each student: this includes past SAT scores, learning style, best studying time, and more.

Built With

• css
• html
• javascript
• json
• local-ai
• ollama
• os
• python
• qwen3
• streamlit