![\"Keynote](\"/blog/pst2025-industry-summit/img-2.jpg\")
\n The first thing teams try when an agent misbehaves is a better prompt. It\nis the cheapest lever and it is almost never the one that was broken.
\nIn production, the failures cluster somewhere else: a tool returns\nsomething the plan did not anticipate, that output re-enters context, and\nthree steps later the agent is confidently doing the wrong thing. No\nprompt change fixes that, because the prompt was never the problem. The\nloop was.
\nBoundaries. A tool-using agent runs real actions against real state:\nfiles, APIs, code. If the only thing standing between a mis-planned call\nand your filesystem is a sentence in a system prompt, you do not have a\nboundary. You have a suggestion.
\nObservability. If you cannot replay why the agent did what it did,\nstep by step, you are not debugging. You are guessing with extra latency.
\nLoop integrity. The reflection step is the highest-leverage and\nleast-watched part of most agent systems. It is where a bad tool result\ndoes the most damage and where teams instrument the least.
\nI treat the agent as a distributed system whose inputs I don't trust\nblindly. Validate at the execution boundary, make every refusal a\nstructured event the agent can reason about, and instrument the loop before\ntuning the prose. The prompt is the last thing I touch, not the first.
","date_published":"2026-04-22T00:00:00.000Z","tags":["agent-systems","reliability","architecture"]},{"id":"https://www.aghoghomena.com/writing/pst2025-industry-summit","url":"https://www.aghoghomena.com/writing/pst2025-industry-summit","title":"PST2025 Industry Summit: reflections and key takeaways","summary":"Field notes from the 2025 Privacy, Security and Trust Industry Summit: AI as a threat amplifier, cyber attribution, and agentic-AI risk frameworks from TD Bank, Mastercard, and the CIC.","content_html":"Proud to have attended the 2025 Privacy, Security and Trust Industry Summit at the Fredericton Convention Centre, hosted by the Canadian Institute for Cybersecurity at University of New Brunswick. Incredible lineup of keynote speakers included leading organizations like TD Bank, Scotiabank, Sun Life, Mastercard, and more, who all shared critical insights on the evolving digital threat landscape.
\nDisclaimer: The views and opinions expressed here are my own and do not represent those of any organization I am affiliated with. As a graduate researcher and engineer attending the 2025 Privacy, Security and Trust Industry Summit to present our recent publication, my reflections are shared purely from an academic and professional perspective.
\n\n The event opened with a powerful keynote address delivered by Dr. Ali A. Ghorbani, Director of the Canadian Institute for Cybersecurity, accompanied by an outstanding team of interdisciplinary support staff, many of whom he acknowledged repeatedly throughout his remarks and afterwards. His leadership, expertise, and genuine gratitude left a strong impression on all in attendance. Dr. Paul Mazerolle, President and Vice-Chancellor of the University of New Brunswick, followed with a warm welcome that set the stage for distinguished government officials to address the gathering.
\n![\"Honorable](\"/blog/pst2025-industry-summit/img-3.png\")
\n Tricia Geddes, Deputy Minister of Public Safety Canada | Sécurité publique Canada, and the Honorable Rene Legacy, Deputy Premier, Minister of Finance and Treasure Board, Minister of Energy for the Government of New Brunswick, spoke at length on critical issues surrounding cybersecurity, governance, and policy. It was truly inspiring to witness the genuine commitment of these national leaders to strengthening Canada’s cybersecurity landscape. The Honorable Dominic LeBlanc was also scheduled to speak on related matters but was called away to address urgent national issues.
\n![\"Rajiv](\"/blog/pst2025-industry-summit/img-4.jpg\")
\n Rajiv Gupta, Director of the Canadian Cyber Centre, opened with key statistics underscoring the important work being done to support individuals, enterprises, and agencies across Canada. He went on to discuss multi-tiered security architectures used to address incidents, highlighted several unique attacker threat models, and touched on issues of quantum readiness along with the national efforts underway in this area.
\nChris Lynam, Director General of the National Cybercrime Coordination Centre and the Canadian Anti-Fraud Centre (RCMP), together with Colin MacSween, Director General of National Cyber Security at Public Safety Canada | Sécurité publique Canada, provided insights into major operations and campaigns aimed at combating cybercrime. They outlined both independent and joint initiatives and presented provincial statistics that shed light on specific security incidents and trends across the country.
\nJennifer M Sloan, Kostia Nikolaiev, and Igor Opushnyev spoke extensively on critical issues such as cyber attribution, residential proxy detection, and the evolving threat models of emerging AI-powered attacks. Mastercard's strong commitment to advancing cybersecurity excellence was evident as they highlighted key partnerships with the Cyber Attribution Data Center at the University of New Brunswick. During the panel discussions, Kostia Nikolaiev drew on his deep product engineering expertise to unpack complex issues surrounding cyber attribution and related challenges. His insights were consistently practical, strategic, and deeply impactful.
\nThe event featured several nutrition and networking breaks, giving attendees the opportunity to recharge, connect, and engage meaningfully. Special thanks go to Dr. Samita Bai, Ph.D. and Griffin Higgins, who emceed the entire summit and ensured that the appreciation award presentations, along with the overall event schedule, ran seamlessly and efficiently. The outstanding success of the #PST2025 Industry Summit was made possible in no small part by their selfless dedication and tireless efforts.
\n![\"Stephen](\"/blog/pst2025-industry-summit/img-5.jpg\")
\n Stephen Sparkes, Chief Information Security Officer and Pamela E. Simpson, M.A., GCTI, CISSP, AI Business Information Security Officer, representing TD Bank, delivered an in-depth discussion on AI as a threat amplifier. Their session covered pressing issues such as large-scale social engineering, deepfakes, model jailbreaking, supply chain vulnerabilities, and defensive AI strategies including automated red teaming and security testing for AI systems. They also introduced novel architectures for governing AI and its lifecycle, offering strategic recommendations such as implementing a RACI matrix and applying a Capability Maturity Model to track project progress. Importantly, they addressed key Agentic-AI risk frameworks, including the CSA MAESTRO Framework and the NIST AI RMF. Hearing these industry leaders unpack such complex challenges from a real-world, enterprise perspective was both enlightening and thought-provoking, sparking meaningful conversations among attendees.
\n![\"Paul](\"/blog/pst2025-industry-summit/img-6.png\")
\n Paul Hanley, Senior Vice President of Cyber Security at Rogers Communications, opened with a broad overview of today’s cyber risk landscape, highlighting issues such as ransomware (including Fatboy Ransomware), extortion attacks, the growing prevalence of targeted threats, and the role of the dark web and underground marketplaces.
\n![\"A](\"/blog/pst2025-industry-summit/img-7.jpg\")
\n He then explored emerging trends in cybersecurity, including:
\nHis depth of expertise was evident throughout, as he skillfully combined technical insight with humor to make increasingly complex security issues and trends both accessible and engaging.
\n![\"Panel](\"/blog/pst2025-industry-summit/img-8.jpg\")
\n Dr. Argyri Panezi, Canada Research Chair in Digital Information Law and Policy at the University of New Brunswick, addressed critical issues surrounding cyber attribution, its various classifications, and its inherently interdisciplinary nature. Her expertise as a seasoned legal scholar proved invaluable, particularly during the panel discussions, where she used clear legal analogies to unpack complex questions, ensuring they were thoroughly examined and accessible to the diverse international audience.
\nKelly Anderson, Director of International Cyber and Critical Technology Policy at Global Affairs Canada, delivered insightful remarks on key diplomatic challenges. She skillfully unpacked the complexities of cyber attribution, highlighting its profound implications for political, technical, and international actors alike.
\nDr. Kwasi Boakye-Boateng, Deputy Director of Research and Training at the Cyber Attribution Data Center (CADC), presented the groundbreaking work being undertaken by the Canadian Institute for Cybersecurity to address industry challenges in cyber attribution. He highlighted the use of deceptive networking systems, internet monitoring, and expert threat intelligence, and introduced a comprehensive framework while also candidly discussing key challenges in the field.
\n![\"Dr.](\"/blog/pst2025-industry-summit/img-9.jpg\")
\n Dr. Ahmed Al-Rawi also contributed significantly to the panel, speaking at length on a range of attribution-related issues. His extensive use of practical examples and references to key sources resonated strongly with attendees, bringing clarity and depth to complex debates and striking at the heart of attribution challenges across multiple levels.
\nElaine Hum, Director of Cybersecurity Partnerships at Scotiabank, shared her personal journey in cybersecurity and the unique path she forged through her work alongside CISOs. Her story was both motivating and memorable, culminating in a lighthearted moment where she donned a superhero cape, sparking laughter and smiles while underscoring the power of leadership and mentorship in the field. The same spirit of inspiration and practical insight mirrored the energy I witnessed earlier this year during her presentation at the Federation of Security Professionals’ Security Conference, hosted by RBC.
\nDan Doran, Vice President of Business Development and Marketing at ADGA Group, delivered an insightful talk on the importance of engaging top-level business executives through carefully tailored executive language and communication techniques.
\nAbhay Raman, Senior Vice President and Chief Security Officer at Sun Life, addressed targeted risk management challenges and proactive security strategies, with particular focus on emerging issues in Agentic-AI powered systems.
\n![\"Dr.](\"/blog/pst2025-industry-summit/img-10.jpg\")
\n As part of the summit, distinguished researchers from academia showcased posters highlighting unique projects being conducted across various universities and faculties. These presentations were enriched by engagement with guests, visiting scholars, and fellow presenters, sparking meaningful discussions around innovative research. To conclude, an awards ceremony was held where four research groups were recognized for their outstanding contributions and achievements.
\n![\"A](\"/blog/pst2025-industry-summit/img-11.jpg\")
\n It was an absolute honor to attend the #PST2025 Industry Summit. I thoroughly enjoyed the event, from the insightful keynotes and panel discussions to the vibrant networking sessions that connected industry, academia, and government leaders.
\nBy way of introduction, my name is Aghoghomena Akasukpe, a systems & full-stack engineer and graduate researcher at Ontario Tech University. I build the infrastructure under AI agents and ship full-stack product, and my research adds rigor to that work. I had the privilege of presenting our accepted paper, “A Per-Bag Suspicion-Based Bagging Strategy for Fighting Poisoning Attacks in Classification”, which proposes novel defense strategies against data poisoning attacks in machine learning systems.
\n![\"Aghoghomena](\"/blog/pst2025-industry-summit/img-12.jpg\")
\n The summit was an incredibly inspiring experience, and I leave even more motivated to contribute to advancing cybersecurity and AI safety. My sincerest gratitude goes to the organizers, speakers, and fellow researchers who made the event impactful and memorable.
","date_published":"2025-08-31T00:00:00.000Z","tags":["research","conference","agentic-ai"]},{"id":"https://www.aghoghomena.com/writing/role-based-access-control-nextjs-middleware","url":"https://www.aghoghomena.com/writing/role-based-access-control-nextjs-middleware","title":"Role-based access control using Next.js middleware","summary":"A walkthrough for enforcing role-based page access in a multi-role Next.js App Router app: JWT decoding in middleware, whitelists per role, and redirect logic with the full code.","content_html":"In this article we will explore an approach to implementing Role-Based Access control using Middlewares when multiple user types exist in a Next.js application.
\nIdeally, you should be familiar with the following:
\nSetup 1: A frontend system must integrate with an Application Programming Interface(API)💻 that has multiple User Roles.
\nSetup 2: An authentication system using JSON WEB TOKENS(JWT's) or an OAUTH2.0 scheme.
\nGoal: A robust access control system (Authorization-after-Authentication) where access is granted strictly based on a users role.
-> What this means?
\nWhat is Role-Based Access Control?
\nRole-Based Access Control (RBAC) is a method of restricting network access(in our case Page📃 Access) based on the roles of individual users within an organization. Learn More
\nWhat are JSON Web Tokens?
\nJWT (JSON Web Token) is a compact, URL-safe token format that can be used for authentication in web applications. Learn More
\nTypical workflow:
\nNext.JS Middlewares
\nNext.js middleware allows us to run code before a request is completed. Then, based on the incoming request, we can modify the response by rewriting, redirecting, modifying the (request or response headers), or (responding directly). Learn More
\nYou should have following to follow with this walkthrough:
\nFirstly, we initialize an empty Next.js project:
\nnpx create-next-app@14 rbaccessnext\nAccept all the default configurations.
\n![\"Output](\"/blog/role-based-access-control-nextjs-middleware/img-2.png\")
\n Voila, now let's run our created project:
\ncd rbaccessnext\nnpm run dev\nWe have our beautiful default landing page:
\n![\"Default](\"/blog/role-based-access-control-nextjs-middleware/img-3.png\")
\n Let's clean up this boiler plate:
\n.tsx files from /app![\"After](\"/blog/role-based-access-control-nextjs-middleware/img-4.png\")
\n ![\"Output](\"/blog/role-based-access-control-nextjs-middleware/img-5.png\")
\n Since we are using App Router, given that we installed Next.js version 14:
\n![\"File](\"/blog/role-based-access-control-nextjs-middleware/img-6.png\")
\n ![\"Home](\"/blog/role-based-access-control-nextjs-middleware/img-7.png\")
\n ![\"Output](\"/blog/role-based-access-control-nextjs-middleware/img-8.png\")
\n Firstly, we create a middleware.ts file in the src diretory
\n![\"Created](\"/blog/role-based-access-control-nextjs-middleware/img-9.png\")
\n Next we add some necessary packages to work with JWT's and Cookies in Next.js.
\nThe packages are
\nnpm i cookies-next dayjs jwt-decode\nWe then advance to importing some required packages:
\n![\"middleware.ts](\"/blog/role-based-access-control-nextjs-middleware/img-10.png\")
\n Next we define some types in TypeScript:
\n![\"Role-based](\"/blog/role-based-access-control-nextjs-middleware/img-11.png\")
\nNext we define some variables (loginURLS, teacherURLS, studentURLS, protectedURLS) at the Top level to help with page restrictions
\n![\"Role-based](\"/blog/role-based-access-control-nextjs-middleware/img-12.png\")
\nFinally, we get to defining the actual middleware😤, the typical structure for middlewares in NextJS is
\nexport default function middleware(req: NextRequest) {\n const res = NextResponse.next();\n\n return res;\n}\nThe following middeware logic below from which we will build on handles:
\n![\"Role-based](\"/blog/role-based-access-control-nextjs-middleware/img-13.png\")
\nAfter this, we define two functions below to do the following:
\n![\"Role-based](\"/blog/role-based-access-control-nextjs-middleware/img-14.png\")
\nLastly, we handle the actual redirection of users based on the isAuthenticated variable and other things defined above. The below image does the following:
\n![\"Role-based](\"/blog/role-based-access-control-nextjs-middleware/img-15.png\")
\nYayy🍃, we are done.
\nThe completed code is attached below, brace yourself:)
\nimport { deleteCookie, hasCookie } from \"cookies-next\";\nimport { NextResponse } from \"next/server\";\nimport { NextRequest } from \"next/server\";\nimport { cookies } from \"next/headers\";\nimport dayjs from \"dayjs\";\nimport { jwtDecode } from \"jwt-decode\";\n\nexport type UserEnumTypes = \"STUDENT\" | \"TEACHER\" | \"ADMIN\"; // Add more User Roles at will\n\nexport type Tokens = { // Actual Token Type\n access: string;\n refresh?: string;\n}\n\nexport type UserType = { // Payload when decrypted from JWT\n accountType: UserEnumTypes;\n accountid: string;\n exp: number;\n firstName?: string;\n iat: number;\n lastName?: string;\n updatedAt: string;\n}\n\nconst loginUrls = [ // Re-route after USER is authenticated\n \"/login\",\n // Add routes that should not be accessible when logged in here\n];\n\nconst studentUrls = [ // URLS meant for students only\n \"/student\",\n // Add more protected routes here\n];\n\nconst teacherUrls = [ // URLS meant for teachers only\n \"/teacher\",\n // Add more protected routes here\n];\n\nconst protectedRoutes = [...studentUrls, ...teacherUrls]; // Add all protected URLS here\n\nexport default function middleware(req: NextRequest) {\n const res = NextResponse.next();\n let isAuthenticated = false;\n let userType: UserEnumTypes = \"STUDENT\"; // Default\n const BASE_FRONTEND_URL = req.nextUrl.origin;\n const CURRENT_URL_PATHNAME = req.nextUrl.pathname;\n\n // First page a user will see(Redirected to) when they login based on role\n const DASHBOARDS = {\n student: \"/student\",\n teacher: \"/teacher\",\n // Can and might be a dashboard page or any other\n }\n\n // Checking that cookie exists and handling it\n if (hasCookie(\"authTokens\", { cookies })) {\n let authTokens = req.cookies.get('authTokens')?.value // Get cookies using Key(authTokens)\n\n if (authTokens && authTokens !== null && authTokens !== \"{}\") { // Checking for empty cookies\n const tokens = JSON.parse(authTokens) as Tokens; // Casting to actual Token Type\n const accessToken = tokens?.access; // Getting the access token\n\n if (accessToken) {\n // Decoding Payload from access Token\n const data = jwtDecode(accessToken) as UserType;\n\n // Checking for Token expiry\n const isExpired = dayjs.unix(data.exp as number).diff(dayjs()) < 1;\n\n // If token is not expired set (isAuthenticated to True)\n if (!isExpired) {\n userType = data.accountType as UserEnumTypes;\n isAuthenticated = true;\n } else {\n // Logout user if Token has expired\n // (Deleted three times because I have trust issues)\n req.cookies.delete(\"authTokens\");\n deleteCookie(\"authTokens\");\n cookies().delete(\"authTokens\");\n }\n }\n }\n }\n\n // Check if the current path has a matching route\n /**\n * E.g\n * currentRoute == \"/student/remarks\"\n */\n const hasRoute = (routes: Array<string>, currentPath: string) => {\n let isValid = false;\n\n routes.map(route => {\n if (currentPath.includes(route)) {\n isValid = true;\n }\n });\n\n return isValid;\n }\n\n // Builds URL to be returned\n const buildUrl = (route: string) => {\n const absoluteURL = new URL(route, BASE_FRONTEND_URL);\n return absoluteURL.toString();\n }\n\n // IF USER IS UNATHENTICATED AND TRIES TO access a PROTECTED ROUTE (REDIRECTION)\n if (!isAuthenticated && hasRoute(protectedRoutes, CURRENT_URL_PATHNAME)) {\n return NextResponse.redirect(buildUrl(\"/login\")); // SWAP FOR YOUR BASE LOGIN URL\n }\n\n // Access control based on user type for protected routes:\n // - Check if the user is authenticated\n // - Check if the current path matches the protected routes\n // - Redirect based on user type:\n // - If userType is STUDENT and the current path is not in studentUrls, redirect to student dashboard\n // - If userType is TEACHER and the current path is not in teacherUrls, redirect to teacher dashboard\n if (isAuthenticated && hasRoute(protectedRoutes, CURRENT_URL_PATHNAME)) {\n switch (userType) {\n case \"TEACHER\":\n if (!hasRoute(teacherUrls, CURRENT_URL_PATHNAME)) {\n return NextResponse.redirect(buildUrl(DASHBOARDS.teacher));\n }\n break;\n case \"STUDENT\":\n if (!hasRoute(studentUrls, CURRENT_URL_PATHNAME)) {\n return NextResponse.redirect(buildUrl(DASHBOARDS.student));\n }\n break;\n\n // Add more -CASES- as you need\n default: // Random fallback if all cases fail (DO AS YOU WISH)\n return NextResponse.redirect(buildUrl(DASHBOARDS.student));\n }\n }\n\n // BLocking access to Login URLS when authenticated\n if (isAuthenticated && hasRoute(loginUrls, CURRENT_URL_PATHNAME)) {\n if (userType === \"STUDENT\") {\n return NextResponse.redirect(buildUrl(DASHBOARDS.student));\n } else if (userType === \"TEACHER\" ) {\n return NextResponse.redirect(buildUrl(DASHBOARDS.teacher));\n }\n\n // ADD MORE TYPES AS YOU WISH\n }\n\n return res;\n}\nNow when we visit /student or /teacher we get redirected to /login which does not exist yet.
\nWe will be using the following hardcoded tokens for authentication for simplicity, but you can integrate with any Backend API for this (Just verify the payload structure and adapt the code to fit).
\nThe Student JWT we will be using:
\n{\"access\":\"eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJhY2NvdW50aWQiOiJmYmFhMDU0OS05YTJhLTQwZmEtYTAyNi01MDIyMWYyOGRiZDciLCJmaXJzdE5hbWUiOiJqb2huIiwibGFzdE5hbWUiOiJzdGFtb3MiLCJ1cGRhdGVkQXQiOiIyMDI0LTA0LTI0VDE2OjE4OjA5LjM3OFoiLCJhY2NvdW50VHlwZSI6IlNUVURFTlQiLCJpYXQiOjE3MTQ5MTY0MzgsImV4cCI6MTcxNDkzMDgzOH0.7XIWJ7bqQWEhLEKDsHr0MQgnVEKVaRCl4IhfQh67HFw\"}\nWhen we decrypt the Payload (using jwt.io) we get:
\n![\"Decoded](\"/blog/role-based-access-control-nextjs-middleware/img-16.png\")
\n The Teacher JWT we will be using:
\n{\"access\":\"eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJhY2NvdW50aWQiOiIwNDUwMWU1Ni0wODE0LTQyNjctYmI4Yy03Nzk2YmZjOGQ5MTIiLCJmaXJzdE5hbWUiOiJVbml2ZXJzYWwiLCJsYXN0TmFtZSI6Ikh1bWFuIiwidXBkYXRlZEF0IjoiMjAyNC0wNC0yNFQxNjoxOToxNy45MDlaIiwiYWNjb3VudFR5cGUiOiJURUFDSEVSIiwiaWF0IjoxNzE0OTE2NjE0LCJleHAiOjE3MTQ5MzEwMTR9.emIvZBIdrRK2W_9UGnsJtbIdlpbMLse6_fYMUrkRPgw\"}\nWhen we decrypt the Payload (using jwt.io) we get:
\n![\"Decoded](\"/blog/role-based-access-control-nextjs-middleware/img-17.png\")
\n Note: The accountType in these payload matches the UserEnumTypes we defined in the middelware.ts file
\n![\"TypeScript](\"/blog/role-based-access-control-nextjs-middleware/img-18.png\")
\n Now let's create a folder /login and a page.tsx inside
\n![\"Initial](\"/blog/role-based-access-control-nextjs-middleware/img-19.png\")
\n Now when we visit /student or /teacher we get redirected to /login which has:
\n![\"Initial](\"/blog/role-based-access-control-nextjs-middleware/img-20.png\")
\n Our final user interface will look like:
\n![\"Final](\"/blog/role-based-access-control-nextjs-middleware/img-21.png\")
\n The code that designs and handles the final interface is attached below:
\n![\"Final](\"/blog/role-based-access-control-nextjs-middleware/img-22.png\")
\n Now when we click student and login:
\n![\"Role-based](\"/blog/role-based-access-control-nextjs-middleware/img-23.png\")
\nWe get redirected correctly and can access /student:
\n![\"Successful](\"/blog/role-based-access-control-nextjs-middleware/img-24.png\")
\n Redirection successful ✅✅✅✅.
\nNote:
\n\n\nWe successfully restricted page access when users authenticate based on their roles✅✅.
\n
View the completed Github Repository.
","date_published":"2024-05-05T00:00:00.000Z","tags":["nextjs","authentication","web-engineering"]},{"id":"https://www.aghoghomena.com/writing/content-based-recommendation-django","url":"https://www.aghoghomena.com/writing/content-based-recommendation-django","title":"Content-based recommendation system with Django and scikit-learn","summary":"Building a content-based recommendation API with Django, Django REST Framework, TF-IDF vectorization, and cosine similarity, with the trained model cached via pickle on app startup.","content_html":"In this article we will explore how we can utilize Django, Django Rest framework TF-IDF Vectorization, Scikit-learn and Cosine Similarity, to build a fully functional Recommendation system.
\nPropsed Recommendation model:
\n![\"Proposed](\"/blog/content-based-recommendation-django/img-2.jpg\")
\n Tools to be used:
\nSide Note 1: Link to Github for the completed project - Here
\nSide Note 2: Link to research paper - Here
\nYou should have the following setup:
\nFirstly we setup our Virtual environment using
\n![\"Setting](\"/blog/content-based-recommendation-django/img-3.png\")
\n Below is the requirements.txt
\nasgiref==3.6.0\ncachetools==4.2.4\ncertifi==2022.12.7\ncharset-normalizer==3.1.0\ndj-database-url==1.2.0\nDjango==4.1.7\ndjango-cors-headers==3.14.0\ndjango-filter==22.1\ndjango-googledrive-storage==1.6.0\ndjango-pandas==0.6.6\ndjangorestframework==3.14.0\ngoogle-api-core==2.10.2\ngoogle-api-python-client==2.86.0\ngoogle-auth==1.35.0\ngoogle-auth-httplib2==0.1.0\ngoogleapis-common-protos==1.59.0\ngunicorn==20.1.0\nhttplib2==0.22.0\nidna==3.4\nitsdangerous==2.1.2\njoblib==1.2.0\nnumpy==1.24.2\npandas==1.5.3\nPillow==9.4.0\nprotobuf==4.22.4\npsycopg2-binary==2.9.5\npyasn1==0.5.0\npyasn1-modules==0.3.0\npyparsing==3.0.9\npython-dateutil==2.8.2\npython-dotenv==1.0.0\npytz==2022.7.1\nrequests==2.30.0\nrsa==4.9\nscikit-learn==1.2.1\nscipy==1.10.1\nsix==1.16.0\nsqlparse==0.4.3\nthreadpoolctl==3.1.0\ntzdata==2023.3\nuritemplate==4.1.1\nurllib3==2.0.2\nwhitenoise==6.4.0\nAfter that we setup Django using:
\ndjango-admin startproject ecommerce\nNext we create a new app using:
\ndjango-admin startapp recommender\nNext we setup the app in our ecommerce/settings.py
![\"settings.py\"](\"/blog/content-based-recommendation-django/img-4.png\")
\n Lets create a recommender/models.py file and include the below content (File Here):
![\"models.py\"](\"/blog/content-based-recommendation-django/img-5.png\")
\n We then register the file in recommender/admin.py
\nfrom django.contrib import admin\nfrom .models import SimilarityModel\n\nadmin.site.register(SimilarityModel)\nNow in recommender/views.py we'll have (File Here)
![\"recommender/views.py\"](\"/blog/content-based-recommendation-django/img-6.png\")
\n Finally for the recommender, we setup recommender/apps.py to ensure we train our models immediately our server is spawned on production (Learn more about ready() and Django Signals).
from django.apps import AppConfig\n\nclass RecommenderConfig(AppConfig):\n default_auto_field = 'django.db.models.BigAutoField'\n name = 'recommender'\n\n def ready(self):\n from .views import train_model_init\n train_model_init() # Handle initial setup of model\nThe end.🔚
\nThis article primarily touched the fundamentals of building a full-scale content-based recommendation systems with Django. View the Completed Project to understand more how the various apps within the Django project communicate and interact.
\n