1786PAL
1786PAL is a course-specific, multi-agent RAG teaching assistant that emulates ECE1786 course staff and adapts tutoring, evaluation, and feedback using course materials and learner history.
ECE 1786 2025 Projects with Summaries
27 projects
AdPersona
AdPersona uses a multi-agent LLM workflow to rewrite e-commerce product advertisements around inferred customer psychological profiles, closing the gap between recommendation accuracy and persuasive presentation.
Bioscribe
BioScribe accelerates traumatic brain injury literature review by filtering papers and extracting structured study, demographic, and cognitive-test data into research-ready datasets.
ChangeTalk
ChangeTalk provides a realistic virtual smoker for motivational-interviewing practice and gives clinicians quantitative feedback on how well their conversation elicits change talk.
ChatIsThisREAL
ChatIsThisREAL analyzes long-form writing sentence by sentence, labels factual claims as real, fake, controversial, or unverifiable, and attaches evidence for human review.
DoctorDash
DoctorDash uses multi-agent medical reasoning to turn symptom descriptions into a likely diagnosis and specialist recommendations while positioning the output as guidance rather than professional medical advice.
EduGenie
EduGenie generates classroom-ready Pre-Calculus lesson packages from a topic, class duration, student level, and teaching style to reduce teacher preparation time.
EmoTuner
EmoTuner creates genre-aware playlists that gradually move listeners from a current emotion to a target emotion through LLM-guided song sequencing.
EventGen AI
EventGen AI recommends Toronto events by converting user preferences and crawled event metadata into comparable LLM-generated embeddings for personalized ranking and explanations.
EVNewsDeconstructor
EVNewsDeconstructor extracts claims from electric-vehicle news articles, checks them against scientific evidence, and produces stance labels plus an article-level credibility score.
FrameTime
FrameTime uses LLMs to generate personalized student timetables that translate tasks, deadlines, and workload constraints into a paced daily schedule.
LeakedIn
LeakedIn assesses privacy and social-engineering risk in LinkedIn profiles and rewrites profile content to preserve professional value while reducing sensitive exposure.
Library Learning Agent
Library Learning Agent tackles 1D-ARC reasoning tasks by building a reusable library of typed natural-language transformation functions instead of solving every puzzle from scratch.
NavigateU
NavigateU maps student goals, completed courses, and target skills to UofT courses and external resources, producing an actionable curriculum plan across STEM career paths.
PlayItOrNot
PlayItOrNot analyzes a Steam user's library, reviews, and a candidate game to produce a persuasive recommendation about whether the user is likely to enjoy and play it.
Plot ScriPT
Plot ScriPT supports script writers by summarizing scripts, tracking characters and story arcs, and predicting plausible future episode developments.
QuickLit
QuickLit automates literature discovery by retrieving and ranking arXiv papers, producing section-level summaries, and supporting grounded chat over the selected corpus.
ReLit
ReLit generates a Related Works section from an abstract or project description by finding relevant prior research and synthesizing how it frames the new work.
SafeMine
SafeMineAI automates coal-mine compliance analysis by matching operation procedures against dispersed regulatory conditions and producing clause-level conclusions.
SiteSage
SiteSage evaluates retail locations from an address and store concept using customer, traffic, and competition signals to return a score with grounded pros and cons.
Slide2Study
Slide2Study turns lecture-slide PDFs into slide summaries and Bloom's Taxonomy practice questions, guiding students through an interactive active-learning loop.
SmarterVision
SmarterVision decomposes complex visual queries into modular perception and reasoning steps so models can answer spatial, relational, and compositional object questions more reliably.
VisMood
VisMood adds trainable emotional modulation to image captioning so generated captions remain visually grounded while expressing controllable emotional tone.
VoicePersona
VoicePersona combines few-shot text-to-speech with LLM-based prosody planning to generate speech that preserves a speaker's identity while expressing a target emotion.
WatchDeal
WatchDeal monitors Amazon.ca products, extracts review-based product aspects, infers user preferences, and sends personalized discount alerts only for relevant deals.
WellSight
WellSight transforms bank transaction exports into categorized spending data, personalized financial insights, practical tips, and dynamic visual summaries.
WhyCombinator
WhyCombinator evaluates YC applications using the written form, founder profile, and intro-video transcript to estimate interview odds and provide actionable improvement feedback.