triAI Cohort 10 · Projects
← Cohort page
🏭 Capstone Projects

Build something
that matters

Nine real-world challenges across language modeling, cultural preservation, healthcare, and more. Choose your track, form your team, and build a Small Language Model that solves a genuine problem.

9Project tracks
Week 4–7Selection window
Week 18Demo Day
Kaggle & CodabenchCompetition platforms
What to expect
Core challenge requirements
📄

No provided datasets

Data engineering is part of the challenge. You are responsible for sourcing, scraping, or generating your own dataset to train and validate your model.

📊

Robust evaluation

You must design and implement a proper evaluation pipeline — just like in real-world engineering. Metrics matter as much as model quality.

👀

Hidden test set

Optimising for the public leaderboard alone is not enough. Final solutions may be stress-tested against a hidden dataset to ensure true generalisability.

💡 Tip: Start thinking about your data collection strategy early. A model is only as good as the data it is trained on.
Available tracks
Choose your project
1
Latent Probing for Toxicity Detection
Codabench Latent Probing LLM

Monitor LLM safety by tapping into internal representations. You will implement a latent probing pipeline using Gemma to detect toxic or harmful user requests — extracting embeddings across layers and training linear classifiers to identify where safety-critical features are encoded. A computationally efficient alternative to full guardrail inference.

How to get started on Codabench

  1. 1Click the competition link above and sign in or create an account using the same email you registered with the cohort.
  2. 2Go to My Submissions and download the starter kit from the Files section.
  3. 3Return to My Submissions and submit your starter kit.
  4. 4Refresh and go to Results — you should appear on the leaderboard.
2
Latent Probing for Mental Health Sentiment Classification
Codabench Latent Probing NLP

Mental health discourse is full of subtle cues that standard sentiment tools miss. Using Gemma's internal embeddings across different layers, you will train lightweight linear classifiers to decode psychological states and distress signals — building a computationally efficient framework for non-invasive emotional monitoring.

How to get started on Codabench

  1. 1Click the competition link above and sign in or create an account using the same email you registered with the cohort.
  2. 2Go to My Submissions and download the starter kit from the Files section.
  3. 3Return to My Submissions and submit your starter kit.
  4. 4Refresh and go to Results — you should appear on the leaderboard.
3
Bridging the Tokenization Gap in African LLMs via SuperBPE
Codabench Tokenization African Languages

Standard tokenizers over-fragment African words into meaningless pieces, inflating compute costs and hurting performance. You will build a localized tokenization pipeline trained on native African corpora, then implement SuperBPE — a hierarchical "superword" merging strategy that preserves semantic units across morphologically rich language families.

How to get started on Codabench

  1. 1Click the competition link above and sign in or create an account using the same email you registered with the cohort.
  2. 2Go to My Submissions and download the starter kit from the Files section.
  3. 3Return to My Submissions and submit your starter kit.
  4. 4Refresh and go to Results — you should appear on the leaderboard.
4
Domain-Specific SLM for Legal Advice
Kaggle SLM Legal

Build a Small Language Model for legal advice and regulatory compliance. You will source and curate data from statutes, case law, and legal documents, then train and refine a model capable of addressing legal queries. Flexibility on architecture, fine-tuning approach, and alignment strategy (including RLHF) is encouraged.

5
Domain-Specific SLM for Agriculture & Climate
Kaggle SLM Agriculture

Train a domain-specific Small Language Model to address real-world challenges in agriculture and climate. You will source and curate a high-quality dataset reflecting meaningful agricultural and climate contexts, then design, train, and refine a model that produces coherent and contextually relevant outputs.

6
Domain-Specific SLM for Medical Advice
Kaggle SLM Healthcare

Build a safety-critical SLM for medical advice and clinical decision support. You will curate data from peer-reviewed journals, clinical guidelines, and anonymized healthcare resources, with strong emphasis on data accuracy, patient privacy, and medical ethics. Guardrail implementation — including fact-checking layers or RAG integration — is strongly encouraged.

7
Domain-Specific SLM for African Folktales
Kaggle SLM Cultural

Preserve and celebrate African storytelling by training an SLM to generate traditional folktales. You will curate data from oral histories, translated anthropological texts, and published African literature. The challenge is capturing cultural nuance, dialect representation, and authentic oral narrative structure — not just fluent text generation.

8
Intelligent Complaint Classification Using Localized Transformer Architectures
Kaggle Classification Transformer

Customer complaint systems are overwhelmed by unstructured text. You will build an end-to-end automated complaint classification pipeline using a transformer architecture (BERT or a distilled variant) — handling colloquial language, spelling errors, and class imbalance, then evaluating with precision, recall, and F1 across multi-class categories.

9
Optimizing RAG Document Retrieval for Agronomic Advice
Kaggle RAG Agriculture

Build the retrieval engine at the heart of a RAG system for smallholder farmers. Given a farmer's question, return the most relevant documents from a knowledge base covering crop diseases, soil management, and climate adaptation. You are scored on nDCG@5 — beat the TF-IDF baseline with BM25, dense embeddings, rerankers, or hybrid approaches. No GPU required.

📧 Can't access Codabench? Fill in the Codabench access help form and we'll sort it out.
Schedule
Project timeline
Wks 4–7
Project kickoff & selection
Choose your project track from the nine available options above.
Week 7
Mentor match
Meet your assigned mentor for the first time.
Week 8
Project specification release
Detailed technical specs are released and the project submission portal opens.
Week 9
Challenge 1 deadline — Problem statement
Develop a values-led problem statement based on your selected project, following the guidance in Course 1.
Wks 9–16
Mentorship & development
Regular weekly meetings with your mentor to refine your solution.
Week 11
Challenge 2 deadline — Data Card
Create a Data Card for the dataset used in your project. Review Course 2 for guidance.
Week 13
Challenge 3 deadline — Impact statement
Create an impact statement card covering your model's benefits and risks. Review Course 3 for guidance.
Week 14
Project submission closes
Final project submission deadline.
Week 15
Challenge 4 deadline — Engagement plan
Design a mini-engagement plan for two stakeholder groups, drawing on Course 4. · Selected teams for Demo Day announced.
Week 18
🎉 Demo Day — October 3, 2026
Final presentations and live demonstrations. Each team presents their problem, dataset, model, ethical considerations, and results. Top teams receive prizes.