# purely.ai > An 8-week live AI engineering cohort for engineers who can code with AI but cannot yet explain, debug or defend what it produces. Under 10 people per cohort. Two real projects, peer review, and 1:1 instructor time. purely.ai teaches engineering judgment for AI-assisted development: how models process context, where they fail, how to read a diff before running it, and when to stop prompting and write the code yourself. Sessions are live every Sunday afternoon for eight weeks, two hours each. The skills are about the loop between you and the AI, not a specific framework, so they keep working when models and tools change. The next cohort starts 6 December 2026. ## Who it's for It is for engineers who already ship with AI but cannot fully account for what they shipped. They can prompt, generate, and reach a working demo, yet stall when the agent touches a file they did not expect, or when a reviewer asks why a line exists. After eight weeks the shift is less about knowing more tools and more about how the work feels: reading a diff before running it, catching the wrong assumption early, and explaining a decision without opening a chat window. Next to developers who lean on the same models without that instinct, the difference shows up in review, where there are fewer surprise diffs, fewer silent regressions, and less time spent repairing code nobody can explain. ## Facts - Format: live online cohort, under 10 people - Sessions: 2 hours every Sunday afternoon, for 8 weeks - Length: 8 weeks plus a demo day - Workload: roughly 6 to 8 hours per week, built to run alongside a full-time job - Projects: two real builds, including one async major project - Included: peer review, 1:1 instructor time on request, recordings of missed sessions, lifetime access to the community - AI tools: subscriptions and API credits are provided if you do not have any - You need: your own laptop and usual editor - Audience: engineers who already prompt and ship with AI. Not a learn-to-code course. - Instructor: Dipanshu Rose, 9 years in AI, ML, NLP and automation. Worked with Gartner and NSE, built BerryViral.com, and mentored dozens of engineers and interns. - Price: INR 15,000, one payment for all eight weeks. No application. - Refund: tell us in the first week and the rest is refunded - Entry: enroll and pay online - Start: 6 December 2026 - Stack: whatever you build on. Web, iOS/Android, game dev, systems, data/ML, backend, embedded, or anything else. ## Curriculum 1. Zero to hero: AI-assisted coding — how we got here, problem formulation, model selection, MCP/tools/skills 2. Build something real with AI: Project 1 — guided build, ship by Friday 3. Making AI trustworthy with tests and automated verification — tests, AI loops, reading diffs, when to interrupt 4. The art of debugging and handling unfamiliar code — peer review, unfamiliar code, first principles, git, 1:1 Q&A 5. Advanced prompting and cost management — prompt iteration, cost, context management, real tools 6. Code you can leave behind: security, reviews, maintenance — AI-made patterns, security, code review, failure modes, legacy code 7. Capstone project: software meets the real world — ambiguous spec, problem formulation, async build 8. Keeping up with AI, without drowning in noise — AI-off review, the jagged frontier, inference, quantization, finetuning 9. Demo day — present your work, live Q&A, what you would ship next ## Outcomes - Explain every line of AI-assisted code with AI off - Spot the bad diff before you run it - Own your work and maintain it with ease and speed - Interrupt or skip AI when it is the wrong tool - Orient in any new codebase within an hour ## What makes it different - Not a workflow: what to do when it breaks - Not a prompt pack: better ways to frame your own problems - Not a demo: judgment that outlasts trending tools - Not a chat that closes with the course: lifetime access to the community ## FAQ - When does the next cohort start? 6 December 2026. The exact time of the first session is sent after enrolment. - How much time does it take per week? Roughly 6 to 8 hours. Sessions are live and the second project builds async through the week, so it fits around a day job. - How big is a cohort, and how much 1:1 time do I get? Under 10 people. 1-on-1 time with the instructor is available on request. - Do I need to pay for AI tools or API credits? They are provided if you do not have an AI subscription. - Do I need to be an experienced engineer? You should already be writing some code with AI. This is not a learn-to-code intro; it is for people who can prompt and ship but want to know what is actually going on. - Is it live or pre-recorded? Live, in small groups, with two real projects, peer review, and 1:1 instructor time. Recordings are available if you miss a session. - What if I miss a live session? A recording is available, plus 1-on-1 time to clear your questions. - What does it cost? INR 15,000, one payment. - What if it is not for me after I start? Tell us in the first week and we will refund the rest. ## Testimonials - Chinmay Virmani, Founder (Chinsco Studio, Game Dev.): "a perspective shift in how I work"; never handed answers, pointed in directions where I could find them myself. - Roshni Yadav, Data Analyst (VMock): "incredibly patient and invested in my learning"; taught the depth of LLM systems and trusted me with meaningful problems. - Anirudh Chebolu, Founding AI engineer (Interact AI): "shaped my engineering fundamentals"; pushed me to reason from first principles. - Vishwesh Keshri, Software Engineer (SonicWall): "a great mentor"; strong AI knowledge, always willing to share experience and explain things better. ## Pages - [Home](https://purely.ai/): what the cohort is, who it is for, curriculum, format, instructor, price, FAQ - [Enroll](https://purely.ai/enroll): enroll and pay online - [News](https://purely.ai/news): short posts on AI engineering - [Models](https://purely.ai/models): comparison table of current AI models by speed, cost and capability ## Contact - Enquiries: support@purely.ai - Instructor: https://www.linkedin.com/in/dipanshurose/