IIT Madras BS · Diploma · BSSE2002
Before you start Tools in Data Science, read this.
What a senior would tell you over chai: what the course really is, how the assignments work, how to get through the ROE, what the projects want, and what you need for the end term. Written after the May 2026 term and checked against the official course page.
This is not an official course page. Dates and rules change every term, so always check tds.s-anand.net and your term's grading document.
- Final score
- 5 partsEach is 20%: GAs, ROE, P1, P2, end term.
- Assignments
- Best 7 countGA0 is only a self-check, no marks. A bad week or two is survivable.
- ROE
- 45 minOnline, open internet, not proctored.
- End term
- In personNo internet. You must attend it to get a grade.
Haven't registered yet? Take GA0 as a self-check (chapter 05). Even with a low score, you can still continue.
The one rule that surprises everyone
In TDS, everything is allowed. Except in the End Term.
- AI
- Copying
- Sharing code
- Group work
- Solvers
- Hacking your way through the questions
- Things other courses would call cheating
In the course's own words: "It's a practical course. Just get it done. 'How' matters less." The end term is the one exception: in person, no internet, no AI, no friends. Hack the questions, not people. Stay out of other students' accounts and IITM's other systems.
The May 2026 term, drawn to scale
13 May → 13 Sep 2026Swipe sideways to see the whole term →
See the yellow strip? From 17 to 22 July, Project 1, GA3, GA4 and GA5 were all open at the same time. Nobody warns you about this. Now you know.
Chapters
What TDS is
Tools in Data Science is a diploma-level course in the IIT Madras BS programme. Most courses teach you the theory. TDS teaches you the tools people actually use at work: the terminal, Git and GitHub, putting an app on the internet, working with AI models, scraping data from websites, and running things on the cloud.
The course content is only a reference. You learn by doing the assignments, not by watching lectures first.It is designed by S Anand (co-founder of Gramener) and taught with instructors and TAs. Here is how the course describes itself:
- AI will teach you. You get challenges and learn on your own, using AI and people around you. Learning by yourself is one of the skills being tested.
- AI will evaluate you. Many answers are checked by a program or judged by an AI model. Those decisions are final.
- It is practical. Unclear problems, messy data, tight deadlines, limited support. Just like a real job.
- It is public. Anyone can read the content and try the assessments. Enrolled IITM students also get Discourse, project evaluation, the end term and a certificate.
"It's a hard course. Take it in your last IITM BS Diploma term. It's good for learning, maybe not for grades."tds.s-anand.net, May 2026
That line is not a joke. In the January 2025 term, 40% of the batch failed. Students still rate what they learnt highly. Go in knowing it will take more time than a normal 3-credit course.
How it differs from other courses
If you have done MLF, MLT or BDM, forget most of what you expect. Nearly everything about TDS works differently.
| Thing | Most IITM courses | TDS |
|---|---|---|
| Where you submit | The SEEK portal | Its own TDS exam portal. SEEK is not used for submissions. |
| Using ChatGPT or Claude | Counts as plagiarism unless allowed | Allowed and encouraged, everywhere except the in-person end term |
| Copying from friends | Honour code violation | Allowed. Share code, work in groups. Understand what you copy. |
| Weekly work | Assignments on a fixed weekly rhythm | Weekly GAs whose windows overlap. Only the best 7 count, and GA0 carries no marks. |
| Quizzes | Quiz 1 and Quiz 2 at exam centres | No quizzes at all |
| Online exam | OPPE: proctored, needs the system test | ROE: 45 min, open internet, not proctored, no system test |
| Your questions | Same for everyone | Generated from your email, so your numbers differ from your friend's |
| Who checks answers | Fixed answer keys | Code that tests your work, or an AI judge. No appeal against the AI. |
| Projects | Varies | Two take-home projects, 20% each |
| Time needed | A normal 3-credit load | More. Plan extra time every week. |
When should you take it?
The course team's answer is your last diploma term. By then you are comfortable with Python and have done enough courses to handle the load. If you have to take it earlier, take GA0 seriously (chapter 05) and keep your other courses light that term.
What you will learn
The May 2026 term had a five-day bootcamp and then eight weekly modules. Don't worry about the tool names. Nobody knows these on day one. That is what the week is for.
- Bootcamp5 days
Set your laptop up properly: the terminal, VS Code, running Python projects, how websites talk to each other, your browser's developer tools, and Git.
terminal · VS Code · uv · HTTP · DevTools · curl · Git · GitHub
- Week 1Development tools
The everyday toolkit: editor, shell scripts, version control, a small database, data formats, and hosting a page for free.
VS Code · uv · Bash · Git & GitHub · SQLite · JSON/CSV · GitHub Pages · LaTeX
- Week 2Deployment
Turn your code into something other people can use through a link, and run AI models on your own machine.
FastAPI · CORS · Google login · Docker · hosting platforms · logging · Cloudflare tunnels · Ollama
- Week 3Working with LLMs
Getting reliable output from AI models: good prompts, the right context, structured answers, and finding similar text using embeddings.
prompt engineering · context engineering · structured output · embeddings · LLM CLI tools · AI coding assistants
- Week 4RAG
Making AI answer from your own documents: splitting documents, searching them well, and checking whether the answers are any good.
vector databases · chunking · hybrid search · reranking · GraphRAG · RAGAS
- Week 5AI agents
AI that uses tools and takes actions, and how to keep it safe: limits, guardrails, sandboxes.
tool calling · agent memory · multi-agent systems · MCP · sandboxing
- Week 6Getting data
Collecting data from websites, PDFs, images, audio and video, ethically, and finding things with smart search.
scraping · hidden APIs · Playwright · DuckDB · document parsing · speech & vision · Google dorking · OSINT
- Week 7CI/CD, security, cloud
Automating releases, attacking and defending AI apps, and the basics of running things on cloud servers.
GitHub Actions · Docker · LLM security · OWASP LLM Top 10 · Cloudflare · SSH · serverless · Terraform
- Week 8ML on the cloud
Training, tracking, shrinking and publishing machine-learning models.
Cloud Storage · BigQuery ML · MLflow · fine-tuning · Hugging Face · quantization · model cards
The skills you really walk away with
This list is the real syllabus. The tools change every term. These don't.- Learning a tool you have never heard of, in an evening, with AI helping you.
- Reading a long task description carefully and doing exactly what it asks.
- Putting something on the internet that a stranger's program can open and test.
- Staying calm when the checker says 0, and working out why.
- Working with classmates instead of alone.
Skills to have first
The official list is short and serious: Python, JavaScript, HTML, APIs, Excel, ChatGPT, and data science basics. The course says you need a good understanding, not a passing one. Tick what you can honestly do today.
Am I ready?
0 of 10Tick the boxes above.
Your setup
- A laptop or desktop. A phone is not enough for this course.
- Permission to install software. A locked office laptop will cause pain. The course publishes a list of software and websites you need access to.
- Windows is fine. The bootcamp sets up a Linux terminal inside Windows (WSL), which is what most instructions assume.
- Steady internet, plus a phone hotspot as backup.
- Cloud accounts early. Some tasks need accounts such as Google Cloud, which can ask for card verification before you can use them. Sort this out in week 1, not on deadline day.
GA0 and eligibility
GA0 is a self-check exam. The course site calls it the entrance exam, and it checks whether you have the pre-requisites. It opens before the term starts (13 May in the May 2026 term) and stayed open until 21 June, alongside a five-day bootcamp.
GA0 is not a gate. There is no pass mark you must clear. Skip it or score low, and you can still move ahead with the course. The course team's advice is that if you score below 40%, you should think about taking TDS in a later term. That is advice, not a rule.
Take GA0 before registration anyway. It's the quickest honest check of where you stand.What's in it
In May 2026, GA0 had 25 short tasks covering the basics you will use all term. A few of them, to give you the flavour:
- Move and rename files, and replace text across many files
- Use Git, GitHub and a GitHub Action
- Use your browser's developer tools to find something on a page
- Write SQL, sort and filter a JSON file, calculate a variance
- Run a small web server, and call an AI model to classify sentiment
- "Get an LLM to say Yes", which is a prompt-writing puzzle
GA0 marks don't count. GA1 to GA5 do.
- GA0 adds nothing to your final score. In May 2026 its marks were not added anywhere, not to your GA average and not to end-term eligibility. It is only a check for you, so a low score costs you nothing but the warning.
- The first GAs that matter are GA1 to GA5. To sit the end term, the average of your best 4 of the first 5 GAs must be at least 40. With GA0 out, those five are GA1 to GA5. Don't skip them.
Eligibility check
best 4 of GA1–GA5Type your scores out of 100. Leave a box empty if that GA isn't out yet.
GA0 doubts
Which email should I log in with?
Your IITM student email. Your questions are generated from your email and your marks are saved against it. Logging in with a personal Gmail means your work goes to the wrong place.
How many times can I try?
You can press Check as often as you like before the deadline. What counts is your last saved submission, so press Save after every improvement.
Does GA0 count in my marks?
No. In May 2026, GA0's marks were not added to the GA average or to end-term eligibility. It's a self-check. Take it anyway, because it tells you honestly whether you're ready.
I got 35% (or skipped GA0). Can I still continue?
Yes. GA0 is a self-check, not a gate, and nothing stops you from moving ahead. The course team does advise students below 40% to think about a later term, so be honest about why you scored low. If you ran out of time but understood the tasks once you sat down with them, you'll be fine. If you didn't understand what the questions were asking, use the bootcamp to fix that before GA1.
GA0 looks too hard. Is the whole course like this?
GA0 covers the basics. The weekly GAs go further (deploying apps, AI agents, cloud). The difference is that you'll have a week of practice before each one. If GA0 feels impossible even with AI helping you, that tells you something.
I joined late. Will GA0's deadline be extended?
Don't plan around it. The May 2026 grading document said plainly that assignment deadlines would not be changed.
What is the bootcamp and do I have to attend?
It is a five-day bridge course: laptop setup, terminal basics, VS Code and Python projects, HTTP and browser tools, and Git. Each day has a checklist on the course site. If you already know these, skim the checklists. If not, do all five days. The whole term rests on them.
How to do a GA
A GA (graded assignment) is one web page with a list of questions, each worth some marks. Every question has a Check button that tells you your score. At the bottom there is a Save button.
Check only tells you your score. Save is what records it.Your last saved submission is the one that gets evaluated.
The three kinds of questions
1. Work it out
Download a file or read some data, work something out, paste the answer: a number, a short code, a piece of JSON. If you follow the rule exactly, the marks are yours.
2. Build it and put it online
Make a small app, put it on the internet, and paste its link. The checker visits your link and tests it. Most of the deadline-night panic comes from these.
3. Write it well
A prompt, a short note, an explanation. An AI model reads it and gives marks. Be clear and follow the requested structure.
4. Something unusual
Find where a photo was taken, record a terminal session, collaborate with classmates. Read these twice. They are often simpler than they look.
A routine that works
- Open the GA the day it is released. Read every question before solving any. Note down the marks.
- Quick wins first. High marks, low effort. Leave the monsters for later.
- Open every hidden section. Questions often have collapsed blocks with the exact rules and formats. Missing one is the most common way to lose marks.
- Use AI to understand, not only to answer. Each question has an Ask AI button. Ask it to explain the task in simple words, then solve it step by step.
- Copy the method, not the answer. Your data comes from your email. Your friend's final answer won't match yours, but their approach will work.
- Check, fix, Check again, then Save. Save every time your score goes up.
- Test "put it online" answers like a stranger. Open your link in an incognito window before pressing Check. Free hosting sleeps when idle, so open it first to wake it up.
- Stuck for 30 minutes? Stop. Ask on Discourse, read the question's public GitHub discussion, or ask a friend how they approached it.
Mistakes that cost people marks
- Pressing Check, seeing full marks, closing the tab, and never pressing Save.
- Skipping GA1 to GA5 because "only the best 7 count". Those five decide your end-term eligibility.
- Pasting a URL that works when you're logged in but not for a stranger.
- "Improving" a value the question wrote down exactly. If the question gives a specific word or format, use it as written.
- Starting two days before the deadline, when three other things are due that week.
The nine GAs of May 2026
Example question titles are real, to show what each week feels like. Titles change every term, but the themes are a good preview.
| GA | Theme | Example questions | Open |
|---|---|---|---|
| GA0 | Entrance: the basics no marks | Move and rename files · SQL: average salary by department · Get an LLM to say Yes | 13 May – 21 Jun |
| GA1 | Development tools | Git time travel · Host your portfolio on GitHub Pages · Debug and improve a failing prompt | 17 – 28 Jun |
| GA2 | Deployment | Deploy a CORS-aware metrics API · Expose a local LLM through a tunnel · Multi-container app with Compose | 28 Jun – 8 Jul |
| GA3 | LLM engineering | Invoice data extraction API · Semantic search ranking · Spin up an LLM command-line tool | 5 – 22 Jul |
| GA4 | RAG | Chunking and hybrid retrieval · A RAG evaluation harness · Reciprocal rank fusion by hand | 8 – 22 Jul |
| GA5 | AI agents | Build a live MCP server · Guardrail red-team round-trip · Build a safe AI mailroom agent | 17 Jul – 10 Aug |
| GA6 | Getting data | Scrape Books to Scrape · Crawl a site respecting robots.txt · Decode a hidden signal in audio | 27 Jul – 9 Aug |
| GA7 | CI/CD, security, cloud | Audit a GitHub Actions workflow · An LLM action firewall · Street View: where is this? | 9 – 21 Aug |
| GA8 | ML on the cloud | A leakage-safe BigQuery ML experiment · Promote the right MLflow model · Publish a model card | 18 – 28 Aug |
Guides and solvers from last term, by Bharat and Angad, are in chapter 11.
The ROE, and staying safe
The Remote Online Exam is worth 20%. It lasts 45 minutes, you take it from home, the internet is open, and it is not proctored. No system compatibility test is needed. In May 2026 it was on Sunday 2 August, 1:00 to 1:45 PM.
"The ROE is intentionally impossible."tds.s-anand.net
Nobody is expected to finish it. It tests how you choose under pressure: what to attempt, what to skip, and how fast you can get AI to do the heavy lifting.
What it looked like
The January 2026 ROE had 15 questions. Last term's (May 2026) had 12, and they were all over the place:
- Find it: work out exactly where a Street View photo was taken.
- Untangle data: a ledger full of look-alike characters, a route through an incident map, a simulation of how a web cache behaves.
- Write and judge: a prompt that detects AI-written text, a short memo on where AI could help a business, insights from government telecom survey data.
- Human ones: donate a few marks to classmates, record a short audio clip.
Strategy for the 45 minutes
- First 3 minutes: scan everything. Write the marks next to each question. Don't start solving yet.
- Quick wins first. High marks, low effort. With AI, you won't know what's easy until you try, so try each one briefly.
- Use AI first. Paste the question in. Spend your own brain only on what AI gets wrong.
- The 2-minute rule. Stuck and going nowhere? Move on and come back later.
- Submit the first reasonable answer, then Save. In May 2026 you could re-answer and there was no negative marking. An unsaved right answer is worth zero.
- Last 10 minutes: start nothing new. Save everything, including your guesses.
- Make friends before the exam. Some ROE questions reward working with classmates. The course faculty's own advice after the Jan 2026 ROE: work together, it helps now and later.
Staying safe: the exam's rules
- Log in with your IITM email. Questions and marks are tied to it.
- Internet, AI, notes and friends are allowed in the ROE. The course explicitly allows sharing code in exams, except the in-person end term.
- Your questions are personalised. A friend's final answer won't work for you. Share methods, not values.
- Save regularly. The last saved submission is what counts.
Staying safe: your own setup
More people lose ROE marks to a dead laptop, a logged-out account or a leaked key than to hard questions. Do this the day before.
Machine and internet
- Laptop charged, charger plugged in.
- Phone hotspot ready as backup internet.
- Chrome or Edge. Turn off ad and cookie blockers for the exam site, since login uses third-party cookies.
- Log in to the exam portal once the day before to be sure it works.
- Empty your Downloads folder so exam files are easy to find.
- Mute notifications. Close everything you don't need.
Accounts and keys
- Be logged in to two AI tools, in case one hits a limit.
- If you use API keys or tokens, check they haven't expired. Some student tokens only last about a week.
- Never paste a key into a WhatsApp group, a shared doc or a public GitHub repo. If you leak one, cancel it and make a new one right away.
- Don't run scripts from group chats that you haven't read. A script can quietly send your keys or login cookies to someone else.
- Install Python packages and tools the day before, not during the exam.
Projects
Two take-home projects, 20% each, open internet. They take longer than any GA and they overlap with GAs and the ROE, so start early. The format changes every term. Here is what May 2026 looked like.
Project 1
5 – 30 Jul 2026Five parts, mostly about using AI and the cloud to get real work done:
- 12.5marks
Requirements interview. Record a short interview and share it, with your questions file, in a Google Drive folder.
- 12.5marks
Differentiating model intelligence. Write one prompt that a weaker AI model gets wrong and a stronger one gets right.
- 12.5marks
AI agent: cloud bucket setup. Get an AI agent to create a storage bucket on Google Cloud, and keep the agent's log.
- 12.5marks
AI agent: dataset upload. Upload a given file to that storage, unchanged, and prove it.
- 37.5marks
Data analyst Telegram bot. A bot that answers questions about data using an AI model. The big one.
Project 2
17 Aug – 5 Sep 2026Eight business case studies, like a TV company's month-end numbers, a solar inverter impact claim, a customs mismatch, a spare-parts search. You play an analyst: read the data, emails and procedures, then write a short diagnostic note with:
- a direct judgment in the first lines,
- an evidence table (claim, source, confidence),
- the explanations you ruled out, and why,
- what you still don't know,
- a safe, reversible next step.
Each note had a character limit, and the key line in the question was: "Evidence is scored before conclusion." A well-argued "insufficient evidence" could earn full marks.
What trips people up
- Panicking over a "half score". In P1, some parts showed only part of the marks on Check because the rest is graded after the deadline. That half was often the full visible score.
- Deleting things after submitting. Grading happens after the deadline. Keep buckets, bots, hosting and API keys alive until results.
- Testing only while logged in. The grader is a stranger. Test your links in incognito.
- Treating an email in the case as proof. In P2, test every claim against the data.
- Leaving it to the last week. P1 overlapped with five GAs. P2 overlapped with GA7 and GA8.
End term and your marks
The end term (ET) is the one part of TDS that looks like a normal exam.
- In person, at an exam centre. May 2026: Sunday 13 September.
- No internet, no AI, no friends. The only assessment where copying isn't allowed.
- Worth 20%, and mandatory. You only get a course grade if you attend it.
- You must be eligible: best 4 of GA1 to GA5 averaging at least 40 (see chapter 05).
The official documents don't publish a paper pattern. According to Bharat's ET prep notes, the May 2026 end term was a theory paper: 80 marks in 90 minutes. It had 30 MCQ/MSQ questions (39 marks) and 9 written "applied AI judgment" questions (41 marks) graded by an LLM, with no negative marking. So the course is practical all term, and then the end term tests theory. Prepare for that switch.
How your final score is calculated
T = 0.2 × GA + 0.2 × ROE + 0.2 × P1 + 0.2 × P2 + 0.2 × ETMay 2026 grading document. GA = average of your best 7 GAs. GA0 was not counted.
Marks calculator
scores out of 100Graded assignments (GA0 has no marks)
Everything else
- GA average (best 7)
- –Enter GA scores
- Total T
- –Empty boxes count as 0
- ET eligibility
- Waiting
Grade bands in the BS programme: S at 90 and above, A 80, B 70, C 60, D 50, E 40. Below 40 is U, which means repeating the course. Confirm the cut-offs in the Student Handbook for your term. Nothing you type here is sent anywhere.
What changed in May 2026
If a senior from 2025 is advising you, some of what they remember is out of date. Here is what changed, compared with the course pages of earlier terms.
| What | Before | May 2026 |
|---|---|---|
| Readiness check | GA1 doubled as the pre-requisite check (Sep 2025) | A separate GA0 self-check exam (not a gate) and a 5-day bootcamp |
| GAs that count | Best 4 of 7 (May 2025), best 5 of 8 (Sep 2025) | Best 7 count. The official page says "best 7 of 9", but GA0's marks were not added. |
| Weights | May 2025: GAs 15%, end term 25% | All five parts at 20% |
| Syllabus | Sep 2025: dev tools, deployment, AI coding, LLMs, data sourcing, preparation, analysis, visualisation | Rebuilt around AI: RAG, agents, LLM security, CI/CD and cloud, ML on the cloud. Data preparation, analysis and visualisation are no longer separate modules. |
| How you learn | Written lessons per topic | Challenge-first. The content is "only a topic reference", and every question has an Ask AI button. |
| Offline evaluation | Most scores showed up when you pressed Check | A big part of the marks is graded offline, after the deadline. In P1, the cloud tasks showed only half the marks on Check, and the rest came later from your agent logs. The Telegram bot was tested after the deadline. P2's written sections were judged offline by an LLM. Your Check score is not your final score, so keep everything running until results. |
| Project 2 | Sep 2025: live evaluation in a fixed one-hour slot | Eight written analyst case studies, judged on evidence |
| Virtual TA | A custom GPT answered doubts | Retired. Ask AI yourself, then TAs. |
Resources and links
Free AI credits
Cut this out · shared by Bharat
Free AI credits for Claude Code, Cursor and Codex
Sign up on AgentRouter with a GitHub account and get $175 in credits for models like Claude Opus, GPT and GLM. Bharat's guide walks through every step. Five more providers give up to $100 to $200 each.
Get $175: setup guide More providers
- These are unofficial third-party relays, not Anthropic or OpenAI. Your prompts and code pass through their servers, so keep private data and your real API keys out.
- Claude and GPT come in daily batches. If you see "402 quota exhausted", wait for the next batch or switch to GLM. The credits can stop any time.
- Most providers want a GitHub account older than one year. The sign-up links in the guides are referral links with a bonus.
Official
- Course siteSchedule, content, every GA, project and ROE link. Bookmark it.tds.s-anand.net
- Question discussions on GitHubA public thread for almost every question. Read these when stuck.github.com/sanand0/tools-in-data-science-public/discussions
- Grading document, May 2026IITMThe official rules for weights and eligibility. Look for "Tools in Data Science".docs.google.com
- TDS on DiscourseIITM loginAnnouncements, doubts, and answers from TAs.discourse.onlinedegree.iitm.ac.in
- Bootcamp checklistsDay-by-day list of what you should be able to do. Great for chapter 04.tds.s-anand.net/…/bridge-course-syllabus
- Live session recordingsThe course YouTube channel.youtube.com/@se-lr5ff
- Marks dashboardYour component scores and the official formula. GA and ROE scores appear one to two days after the deadline.tds.s-anand.net/marks-dashboard
- Public announcement groupFor anyone auditing the course without an IITM login.groups.google.com/g/tds-iitm
From the course faculty's blog
- What happened in the Jan 2026 ROEWhich questions students got right, how few used AI, and four lessons. Read before your ROE.s-anand.net/blog/tds-jan-2026-roe
- Why the course is built around questions nowThe thinking behind "AI will teach you".s-anand.net/blog/tools-in-data-science-jan-2026
End term prep
Bharat and Angad on GitHub
Guides, solutions and solvers from the May 2026 term, all with source code.
- BharatGitHubGA0 to GA6, Project 1 and Project 2 work, plus the ET prep site.github.com/HypeMonk
- AngadGitHubGA4, GA5, GA7 and GA8 guides and solvers, and Project 1 and Project 2 guides.github.com/angadseth
Project 2
- Project 2 guideby AngadHow to work each of the eight case studies.angadseth.github.io/tds-p2-guide
- Project 2, all eight casesby BharatOne write-up per case, 1A to 4B.github.com/HypeMonk/TDS-P2
- P2 EvaluatortoolPaste your case note and read it back against the rubric: marks per criterion and what to fix first. Free hosting, so the first load can take a minute.tds-p2-evaluator.onrender.com
Solvers, with source code
Solvers are allowed in TDS. These were built for the May 2026 questions, so they may stop working when the questions change next term. Check every answer before you Save, and remember that no solver helps in the end term.
- GA7 solverby AngadWorks out the offline GA7 answers from your email by running the exam's own question generator, and shows the evidence.
- GA8 solverby AngadDerives GA8 Q8 to Q10 from your email, and ships the service for Q1 to Q7 for you to deploy as your own.
- TDS Portal Solverby GT IndianPick the term and exam, enter your IITM email, and answers are generated locally in your browser. Covers the Jan and May 2026 terms.
Method guides
These explain the approach and the traps, one question at a time. Questions change every term, but the way of thinking carries over. Some are in Hinglish.
- GA4 guide: RAG weekMay 2026How to approach each retrieval question, one file per question.github.com/angadseth/tds-ga4-guide
- GA5 guide: AI agents weekMay 2026Includes the general debugging tips that apply to every "put it online" question.github.com/angadseth/tds-ga5-guide
- GA7 handbook: CI/CD, security, cloudMay 2026The order rules must be applied in, and the traps hidden in the checks.angadseth.github.io/tds-ga7-handbook
- Project 1 guideMay 2026All five parts, with the five rules that apply to everything.github.com/angadseth/tds-p1-q5-guide
Older terms
Doubts, answered
The questions every new batch asks in the first two weeks.
Can I really use ChatGPT, Claude or Gemini?
Yes. The grading document says you may use LLMs and may collaborate on assignments. The course asks you to give credit to the AI and people who helped. The only exception is the in-person end term.
Can I use a solver?
Yes. In TDS everything is allowed except in the end term, and that includes solvers. Two cautions: a solver built for last term may not work once the questions change, and a wrong email gives wrong answers. Check before you Save. Links are in chapter 11.
Is copying from friends allowed? Isn't that cheating?
In TDS it's allowed, even encouraged: "Share code. Even in projects, assignments, and exams (except the final in-person end-term exam)." The reason is that real work is done in teams. In practice your questions are generated from your email, so copying the final answer doesn't work anyway. Copy the method and understand it.
Where do I submit? I can't find TDS assignments on SEEK.
That's expected. TDS uses its own exam portal. All GA, project and ROE links are on tds.s-anand.net. SEEK is not used for TDS submissions.
I pressed Check and got full marks. Am I done?
Not until you press Save. Check only evaluates. Your last saved submission is what counts.
Can I skip some GAs?
Only your best 7 GAs count, so a bad week can be dropped. But GA1 to GA5 decide end-term eligibility, so don't skip those. GA0 carries no marks at all. If you're going to skip, skip a late one.
My friend and I did the same thing but got different answers. Who is wrong?
Probably nobody. Questions are personalised from your email, so your data and your expected answer are different from theirs.
The checker gave me 0 but my app works. Why?
Usually the problem isn't your logic, it's the setup around it: the wrong link, a server that fell asleep, something that only works while you're logged in, or an output format slightly different from what the question wrote. Open your link in incognito, re-read every hidden section of the question, and match the exact format.
The AI grader marked my answer unfairly. Can I appeal?
No. "LLM evaluations are final. No appeal." The skill being tested is writing answers that hold up no matter how the AI reads them: clear structure, the exact sections requested, no vague sentences.
Will deadlines be extended?
Plan as if they won't. Extensions have happened in past terms, but the May 2026 grading document said assignment deadlines would not be changed.
Do I have to pay for AI tools or hosting?
Most tasks can be done with free tiers. For AI, see the free AI credits in chapter 11 ($100 to $200 per provider). Some cloud tasks need an account that asks for billing or card verification, such as Google Cloud Storage in Project 1. Set these up early and watch for any charges.
When will my scores show up?
According to the course's marks dashboard, GA and ROE scores are updated one to two days after the deadline. Project scores can take a week or two, because they take longer to evaluate.
Is the ROE really impossible?
It's designed so that nobody finishes. That's fine. Scan everything, take the quick wins, use AI first, and save as you go. See chapter 07.
Is the ROE proctored? Do I need the system compatibility test?
No and no. The ROE is a non-proctored, open-internet exam, and TDS doesn't need the OPPE system test.
What happens if I miss the end term?
You need to attend the end term to get a grade. If you miss only the end term and completed everything else, you get an "I" (incomplete) and can write only the end term in a later term. Check your term's grading document for the exact options and fees.
What does the end term test?
Theory, even though the rest of the course is practical. It's in person, closed book, no internet. The official documents don't give a paper pattern, but according to Bharat's ET prep notes the May 2026 paper was 80 marks in 90 minutes: 30 MCQ/MSQ questions and 9 written questions graded by an LLM, with no negative marking. Start with TDS ET Prep.
Are there bonus marks?
Sometimes. The course may post bonus activities on Discourse, email, SEEK or the course site. They are given at the course team's discretion, so don't count on them.
Can I do TDS without being enrolled?
Yes. The course is public: anyone can read the content and try the assessments. Only enrolled students get Discourse, project evaluation, the end term and a certificate.
Who should I ask when I'm stuck?
The course's own order: ask AI first (twice). Then TAs, for learning doubts like "how do I approach this assignment?". Instructors only for exceptions, with proof, like wrong marks or an extension request. The faculty for suggestions about next term's content.
How do I keep up with announcements?
Check three places daily: SEEK notifications (the bell icon), your email including spam, and TDS Discourse.
My marks look wrong. What do I do?
Collect proof first: screenshots of your saved submission and the score, with dates. Then contact the instructors through the channel given on the course site. Remember that AI-judged scores can't be appealed.
Should I take TDS and MLP together?
MLP is listed as the co-requisite, so many people do. The bigger question is total load: TDS takes more time than a normal course, so keep that term's other courses manageable.
Nothing matches. Ask on Discourse, and maybe it ends up here next term.