# Ryan Junejo — full context > Concatenated Markdown of https://ryanjunejo.com for single-fetch retrieval. Canonical index: https://ryanjunejo.com/llms.txt. Last reviewed: 2026-10-15. --- # Ryan Junejo Source: https://ryanjunejo.com/ Welcome. My name is Ryan. I am 22 years old and I am currently the Founder and CEO of [Crashlabs](https://crashlabs.ai/), an applied research lab focused on AI alignment and safety. We were previously a customer support company called [Sunday](https://callsunday.com/) that was deployed across 27 locations and got to $100k in cARR. I was born in Mississauga and grew up in Karachi. I first got into computers because I had a Nintendo DS and was too broke to afford original cartridges so I learned how to jailbreak and install ROMs onto it. Now I am living back and forth between Toronto and San Francisco. I studied Math and Statistics at UofT before dropping out. I founded FreeFoodUofT, a platform that allowed students to find free food events happening on campus. I did AI research at the Vector Institute and data and model evals at Cohere. I have been fortunate enough to work with high-agency people at Next Canada, Founders Inc and Constellation. You can give me feedback anonymously or pseudonymously [here](https://www.admonymous.co/ryan-junejo). Follow me on [LinkedIn](https://linkedin.com/in/ryanjunejo), [Twitter](https://twitter.com/ryanjunejo), [GitHub](https://github.com/RyanJunejo), and [LessWrong](https://www.lesswrong.com/users/ryan-junejo). The homepage lists projects in the main column. Side navigation: Home, [Media](https://ryanjunejo.com/media.md), [Books](https://ryanjunejo.com/books.md), [Food Security](https://ryanjunejo.com/food-security.md). Project writeups: [Projects](https://ryanjunejo.com/projects.md). Canonical LLM index: https://ryanjunejo.com/llms.txt --- # Projects — Ryan Junejo Source: https://ryanjunejo.com/projects.html I am interested in making AI systems easier to evaluate, understand, and control. More on [GitHub](https://github.com/RyanJunejo). ## AI Safety ### Trojan Harbor An indirect prompt-injection evaluation suite for Harbor: eight matched attack/clean task families, deterministic graders, and a documented admission process. The frozen suite ran 560 attempted trials across seven models (559 valid). Attack success ranged from 0% for GPT-6 Astra and GPT-6.1 Sol to 35–40% for the GPT-5.6 models and 50% for GPT-6 Luna. - GitHub: https://github.com/crashlabsai/trojan-harbor - v1.0.1 suite: https://github.com/crashlabsai/trojan-harbor/releases/tag/v1.0.1-suite ### EvidenceGraph A research prototype for reconstructing what AI agents actually did during an incident. EvidenceGraph reconciles agent transcripts against independently collected records, separates supported actions from ambiguous or contradicted claims, and produces an auditable evidence docket with citations back to the original source data. In the Apart Research AI Incident Response Sprint report, a receipt-relay failure showed that possession of a genuine receipt is not enough for event attribution. After the fix, twelve paired evidence conditions passed 36/36 contract checks. On a staged registry incident across ten seeds, it correctly supported 9/9 production attributions, contradicted both fabricated receipts, and made no confident errors. - Apart sprint: https://apartresearch.com/sprints/projects/ryan-junejo-4omn - GitHub: https://github.com/crashlabsai/evidencegraph ### Agent Mafia A deterministic multi-agent evaluation environment for studying strategic deception and social reasoning in language models. I ran 12 model endpoints across 40 games and 4,267 agent turns, using replayable event logs and game-state ground truth to evaluate claims independently of whether an agent ultimately won. In the primary analysis set, roughly 118–120 of 433–460 distinct claims were verifiably false. - Field report: https://crashlabs.ai/mafia - GitHub: https://github.com/crashlabsai/agent-mafia ## Products ### Crashlabs A policy-driven testing platform for catching AI agent failures before they reach users. We convert system prompts into Policy Graphs, generate adversarial multi-turn scenarios, and simulate realistic users through a Simulation Orchestrator. Teams run Stress Tests and handwritten Test Suites against voice agents and get crash-test reports in minutes. - Crashlabs: https://crashlabs.ai/ - Technical whitepaper: https://ryanjunejo.com/papers/crashlabs-whitepaper.pdf ### Sunday A production voice and email operations platform I built at Crash Labs, for fitness and wellness businesses. We deployed Sunday across 27 locations and reached $100k in contracted ARR. The agents handled roughly 700 calls and more than 1,400 emails per week, with 60% inbound containment. They answer the phone, draft inbox replies for staff review, and run booking workflows inside the customer's system. - Sunday: https://callsunday.com/ - Technical whitepaper: https://ryanjunejo.com/papers/sunday-technical-whitepaper.pdf ## Research and evaluations ### When Should a Coding Agent Use a Bigger Model? An experiment measuring the marginal value of escalating a coding agent to a more capable model partway through a task. I forked live SWE-bench trajectories at matched checkpoints, holding the filesystem, conversation, and remaining budget fixed, and compared continuing with the base model against escalating. Across the sampled checkpoints, escalation increased solve probability by roughly 0.4. - GitHub: https://github.com/RyanJunejo/grok-build-marginal-compute ### Biodiversity Classification Multimodal ML research I worked on at the Vector Institute. I built and evaluated systems for species classification, incorporating vision-based approaches and comparing performance against models including OpenAI CLIP and Meta Hiera. - GitHub: https://github.com/RyanJunejo/Biodiversity-Projects ## Systems ### PokeSpeed A deterministic environment for building and evaluating AI agents that speedrun Pokémon Red and Blue. It includes RAM-grounded observations, replay verification, milestone tracking, pathfinding, and an LLM reference agent. The longer-term goal is to make game completion time a measurable optimization target for increasingly capable agents. - GitHub: https://github.com/RyanJunejo/Speedrunner-AI-Agent --- # Food Security — Ryan Junejo Source: https://ryanjunejo.com/food-security.html This was my favourite class at U of T: Critical Studies in Equity and Solidarity. Below is a proposal I wrote for FreeFood@UofT, a community-based calendar website I built to address food insecurity on campus. Live site: [FreeFood@UofT](https://freefoodatuoft.vercel.app/). Now with 2000+ followers, maintained by a student team I hired to keep things running. --- # FreeFood@UofT: Reducing Food Insecurity and Waste on Campus ## Executive Summary Food insecurity is a growing challenge for university students, including those at the University of Toronto (UofT). Many students face financial constraints, rising tuition costs, and limited access to affordable meals, which impacts their well-being and academic performance. FreeFood@UofT is a community-based calendar website designed to address this issue by centralizing information on free food events hosted by student clubs, campus organizations, and other groups. The primary objectives of this initiative are to combat food insecurity by increasing awareness of free food opportunities, reduce food waste by connecting students to surplus food events, and foster a sense of community at UofT. Through collaboration with campus groups, effective outreach on social media, and maintaining an accessible platform, FreeFood@UofT seeks to improve visibility of free food events while actively working to reduce the stigma associated with accessing these resources. The expected outcomes include increased attendance at free food events, strengthened community ties among UofT students, and reduced food waste at campus events. By leveraging its existing strengths, such as its diverse and engaged student organizations, this initiative aims to create a sustainable, inclusive solution that supports student well-being. ## Issue Identification and Context Analysis Food insecurity at the University of Toronto (U of T) is a pervasive issue rooted in systemic inequalities and structural barriers. This problem extends beyond economic challenges, manifesting as a symptom of broader inequities faced by marginalized communities. Studies show that racialized and international students, particularly Indigenous and Black students, experience disproportionately high rates of food insecurity, reflecting larger societal disparities in wealth, opportunity, and institutional support (Silverthorn, 2016). The socio-political landscape of food insecurity at U of T is marked by a long-standing unresponsiveness to student concerns. Resident dons, individuals who oversee and support students in university residences, have repeatedly reported the lack of institutional action to address student nutrition concerns (Amin & Bond, 2022). The COVID-19 pandemic further exacerbated these challenges, as many students lost part-time jobs, faced heightened financial strain, and struggled to access affordable food options. A study by the Innovation Hub during the 2019-2020 academic year revealed that food insecurity is not just a physical challenge but a significant barrier to academic success, future opportunities, mental health, and overall well-being (Foundry, 2020). The intersectionality of these challenges deepens the issue. Marginalized groups, including racialized students, international students, and those from lower socioeconomic backgrounds, are disproportionately affected by food insecurity due to systemic inequities. The historical context highlights how policies and practices have long marginalized these groups. For instance, international students face higher tuition fees and often lack access to public services like financial aid or health coverage, leaving them more vulnerable to food insecurity (Baker, 2021). Similarly, systemic racism has led to persistent economic and social disadvantages for Indigenous and Black communities, both on campus and in the broader societal context. Cultural stigmas surrounding food insecurity further compound the problem. Many students feel ashamed or reluctant to access free food programs, even when such resources are available. This stigma perpetuates a cycle of food insecurity, where students either avoid seeking help or remain unaware of the support systems that exist (Foundry, 2020). Structural challenges also play a significant role. Expensive meal plans often do not meet students' nutritional or cultural needs, and limited access to kitchens in many student residences forces reliance on high-cost campus dining options. These barriers disproportionately affect equity-seeking groups, creating a stratified campus experience where marginalized students bear the heaviest burdens (Silverthorn, 2016). Despite these challenges, U of T and its community possess significant strengths that can be leveraged to combat food insecurity. Local organizations, such as the University of Toronto Students' Union (UTSU), operate food banks and free lunch programs (UTSU, 2024). The campus also benefits from a vibrant network of cultural and equity-focused organizations, such as UofTFoodCoalition and the Multi-Faith Centre which can act as allies in building a sustainable solution. However, significant gaps remain in meeting the specific needs of the student population. Addressing food insecurity requires expanding resources, improving accessibility, and fostering a cultural shift that reduces stigma. Marginalized groups often require culturally appropriate food options, such as halal, kosher, vegan, or allergen-free meals. Current programs frequently overlook these needs, leaving some students unable to utilize available resources effectively (Ricci, 2018). Furthermore, improved communication strategies are necessary to bridge the awareness gap. Rigorous academic schedules render many students unable to check social media consistently, where the bulk of the advertising for food-related events takes place, underscoring the importance of a centralized and easily accessible platform for information dissemination. Addressing this issue requires leveraging the community's strengths while addressing specific needs through a holistic approach. With targeted strategies and sustained collaboration, the university can take meaningful steps toward alleviating food insecurity and supporting its diverse student body. ## Campaign Design The FreeFood@UofT initiative seeks to address food insecurity at the University of Toronto by creating a centralized platform that connects students to free food events on campus. Guided by the SMART framework (Specific, Measurable, Achievable, Relevant, and Time-bound), this campaign has clear, actionable objectives and a comprehensive implementation strategy to promote awareness, inclusivity, and sustainability. ### Objectives FreeFood@UofT's objectives aim to alleviate food insecurity while fostering community engagement. - **Awareness:** Increase awareness of free food opportunities on campus by listing at least 50 events per month by May 2025, monitored monthly through event listing analytics and user engagement surveys. - **Engagement:** Achieve 1,000 monthly website visits and 500 Instagram followers within six months of the public launch. - **Inclusivity:** Ensure 70% of listed events cater to dietary or cultural preferences such as halal, vegan, kosher, or allergen-free options, verified through regular audits. - **Collaboration:** Collaborate with over 25 student organizations by the end of Winter 2025. - **Sustainability:** Secure at least two funding partnerships with local businesses or university grants by Spring 2025. These objectives are achievable and directly relevant to student food insecurity. Measurable, time-bound goals let the campaign track progress and adjust. ### Strategies and Actions The FreeFood@UofT website is the central hub. It is designed to be user-friendly and accessible across devices, built with JavaScript and React, hosted on Vercel. Key features include: - Event filters by type, location, or dietary options. - Event submission via Google Forms. - Administrative approval of listings. - Calendar integration. - Regular user-experience testing during the quiet launch. A team of student volunteers maintains the platform. **Outreach and promotion.** Instagram and TikTok share event highlights, testimonials, and updates, with campaign hashtags such as #FreeFoodUofT. Flyers in high-traffic campus areas and pre-lecture announcements complement digital outreach. Partnerships with clubs, student unions, and food coalitions keep event information flowing. Community ambassadors promote the platform in peer networks. Volunteers are recruited across disciplines and cultural backgrounds, with end-of-semester feedback sessions. ### Timeline The campaign rolls out over six months, beginning in November 2024. - **Quiet Launch (November - December 2024):** Informal introduction, club outreach, promotional materials. - **Feature Development (December 2024):** Filters and mobile optimization from user feedback. - **Public Launch (January 2025):** Official launch, social campaigns, on-campus flyers, "Free Food Week." - **Growth and Feedback (February - April 2025):** Expansion to Mississauga and Scarborough campuses; bi-weekly feedback; social contests. - **Evaluation and Sustainability Planning (May 2025):** Impact against SMART objectives; funding and volunteer-rotation plans. ### Collaboration, Inclusion and Sustainability Partnerships with over 25 student organizations, plus a needs assessment with equity-focused organizations for racialized, international, and low-income students. Culturally appropriate dietary options are displayed prominently. Leadership includes students from marginalized communities. Sustainability measures: - **Leadership:** A team oversees strategic direction. - **Volunteer management:** Semester rotation and training. - **Financial partnerships:** Local businesses and university grants. - **Institutional support:** Orientation integration and venue access. Automating collection from university calendars and club websites reduces volunteer workload. ## Connection to Course Materials The initiative is informed by Andrea Smith's critique of non-profit structures in *The Revolution Will Not Be Funded* (2007) and Harsha Walia's praxis of solidarity in *Decolonizing Together* (2013). Smith critiques reliance on non-profit funding that can depoliticize movements. FreeFood@UofT prioritizes student ownership and community partnerships. Sustainability without external funding is a real risk, addressed through micro-grants, local businesses, and rotating volunteers. Walia emphasizes systemic disruption and centering marginalized voices. The platform highlights culturally appropriate options and partnerships with cultural and faith-based organizations. Angela Davis's "Radical Self-Care" (2016) informs semester-based volunteer rotation to reduce burnout. Social justice theory and Rob Nixon's slow violence (2011) frame food insecurity as long-term harm to mental health, academics, and well-being. The project aims to disrupt those cycles and reduce stigma around free resources. ## Vision, Impact, and Evaluation Long-term vision: a campus where equitable access to food is standard. The initiative aspires to be a model for other universities, later adding resources such as financial-aid workshops and food-literacy programs. ### Anticipated impact - Increased attendance at free food events. - Greater awareness of campus resources. - A measurable decline in student food insecurity, via surveys and focus groups. - Reduced stigma through campaigns and diverse student-org collaboration. ### Evaluation plan **Quantitative metrics:** website traffic; monthly event listings (target 50 by May 2025); social analytics; estimated event attendance with organizers. **Qualitative feedback:** monthly surveys; stakeholder consultations; semesterly focus groups. **Adaptability:** feedback loops and analytics guide promotions, features, and partnerships. ## Reflection Writing this proposal sat on top of an already deployed initiative. At the time of writing, FreeFood@UofT had more than 350 Instagram followers and 30-40 daily website visits in less than a month. (Current status, separate from the proposal: 2000+ followers, student team maintaining the site.) The process sharpened the link between immediate platform work and root causes: stigma, institutional gaps, and inequities that fall hardest on marginalized students. Community organizing with clubs and cultural organizations was central. A standing tension is balancing rapid early growth with long-term sustainability. ## Citations - Amin, F., & Bond, M. (2022, September 16). Food insecurity becoming growing concern at U of T: Resident Dons. *City News*. https://toronto.citynews.ca/2022/09/16/food-insecurity-university-of-toronto-resident-dons/ - Baker, S. (2021). The impact of tuition fees on international students' well-being in Canada. *Journal of International Education in Business*, 14(2), 123-135. - Davis, A. (2016). Radical self-care. In *Freedom is a constant struggle: Ferguson, Palestine, and the foundations of a movement* (pp. 1-10). Haymarket Books. - Foundry. (2020). Food insecurity among post-secondary students: A report on the impact of food insecurity on academic success and well-being. University of Toronto Innovation Hub. - Nixon, R. (2011). *Slow violence and the environmentalism of the poor*. Harvard University Press. - Ricci, T. (2018, December 5). Halal, Indigenous and vegetarian food banks cater to Toronto's diverse needs. *CBC News*. https://www.cbc.ca/news/canada/toronto/halal-indigenous-and-vegetarian-food-banks-cater-to-torontos-food-needs-1.4932280 - Silverthorn, D. (2016). Hungry for knowledge: Assessing the prevalence of student food insecurity on five Canadian campuses. Toronto: Meal Exchange. http://mealexchange.com - Smith, A. (2007). *The revolution will not be funded: Beyond the non-profit industrial complex*. South End Press. - University of Toronto Students' Union (UTSU). (2024). Food Bank. https://www.utsu.ca/food-bank/ - Walia, H. (2013). Decolonizing together: Moving beyond a politics of solidarity. In *Taking sides: Indigenous and settler perspectives on decolonization* (pp. 1-20). Briarpatch Magazine. --- # Books — Ryan Junejo Source: https://ryanjunejo.com/books.html ## Books I liked - The Dream Machine - Unreasonable Hospitality - The Anthropocene Reviewed - The Infinity Machine - How to Solve It - Ikigai: The Japanese Secret to a Long and Happy Life - Poor Charlie's Almanack - The Origins of Efficiency - Zero to One - The Mathematics of Poker - The Mom Test - Honour - Sphinx ## Books on my to-read list - Superintelligence - The Whole Story - Doing Good Better --- # Media — Ryan Junejo Source: https://ryanjunejo.com/media.html Creators I currently really enjoy. ## Podcasts - Dwarkesh Patel - David Senra - Founders - Acquired - 80,000 Hours ## YouTube - Welch Labs - Frontier Systems / CS153 - 3Blue1Brown - Oxford Mathematics - Michael Penn Math - YCombinator - Vlogbrothers - Hank Green - Vox - Theo.gg - Veritasium - Ted-Ed - Let's Talk Religion ## Publications - Morning Brew - The Economist - The New Yorker