Zoho drive plan
DevOps Engineer (ZLabs), SASTRA Thanjavur, 9 and 10 October 2026. Apply by 10 AM, Friday 25 September. Written 22 September, 17 days out.
You are eligible. This drive allows 5.5 CGPA and up to 3 standing arrears. Your 6.8 CGPA and one arrear clear both bars. This is one of the few drives where the arrear does not matter, so it deserves full effort.
Reality check
- Zoho's funnel is brutal. Reported numbers from campus drives: 1500 sat the written test, 250 cleared. In another, 600 applied, 32 reached the coding round, 5 reached interviews, 1 got the offer.
- The written round is not MCQ. You write the answer. C output questions on pointers and arrays eliminate most people, and you have never written C for this purpose.
- Round 3 (build a working console app in 3 to 4 hours) eliminates more people than every other round combined.
- 17 days is not enough to become a strong programmer from scratch. It is enough to clear round 1 and give round 2 a real shot, if you do 4 to 5 focused hours every day.
- Your realistic goal: clear the written test, solve 2 of the 5 round-2 problems cleanly, and be able to explain your code. That alone puts you ahead of most of the room.
What the rounds actually are
| Round | Format | What kills people |
| 1. Written | Paper, 60 to 90 min. About 10 C "predict the output" questions plus 10 aptitude, fill in the blank, no options, no negative marking. | Pointer arithmetic, pre/post increment inside printf, static variables, string functions. Aptitude is moderate, not the bottleneck. |
| 2. Basic programming | 3 to 6 problems, 3 hours, any language, company laptop, no internet. Problems solved in order. | Not being able to start. Using library shortcuts. Not explaining your approach when the evaluator walks over. |
| 3. Advanced / machine round | Build a console app (railway reservation, taxi booking, inventory, social media) with 4+ modules over 3 to 4 hours. Requirements change midway. | No design before typing. Monolithic code. Crashes on bad input. |
| 4. Technical interview | 30 to 40 min. Resume projects in depth, OOP, DSA variations, SQL, sometimes puzzles on paper. | Not knowing your own project's design decisions. For ZLabs, weak ML basics. |
| 5. HR | Why Zoho, relocation to Chennai, how long you will stay, salary. | Almost nobody fails here. |
Language decision
Python for rounds 2, 3 and 4. C only for reading output questions in round 1.
The job post lists Python first, and ZLabs is an ML team. Python is the fastest way for a beginner to get working code in 3 hours. You still need to read C because round 1 is C output. You do not need to write C programs.
Install before the drive, on your Mac: Python 3 (already there), VS Code. Zoho asks for Windows, JDK, Eclipse, Turbo C, Dev C++. Ask the placement cell whether a Mac with VS Code is accepted. If not, borrow a Windows laptop for the two days. Do not try to make the 2015 Air run Windows.
Practice site: Zoho Prep Room. Start with its Learn tab, lessons 1 and 2, before anything below.
The 17-day plan
Each day: 1 h C output, 2 h Python problems, 1 h aptitude. Mornings for the hard part. Rest on the day 14 evening only.
Week 1: Sept 22 to 28. Foundations, round 1 and 2 basics
Day 1, Sep 22Apply on the form today. Python: variables, loops, lists, strings, functions. Solve 10 easy problems: reverse string, count vowels, FizzBuzz, factorial, prime check. C: read how printf, arrays and pointers work, GeeksforGeeks "C output" set 1.
Day 2Python: dictionaries, sets, nested loops, 2D lists. Problems: frequency count, anagram check, matrix transpose, pattern printing (diamond, pyramid). C output set 2 and 3, 15 questions, write the answer before checking.
Day 3Strings deep: string compression (aaabb to a3b2), reverse words keeping punctuation, longest palindromic substring by brute force. C: pointers and arrays, *p++ vs (*p)++, a[i] equals *(a+i). Aptitude: percentages, ratios, 20 questions timed.
Day 4Sorting and searching by hand: bubble, selection, binary search. Problem: alternate sorting (max, min, second max, second min), merge two sorted arrays without duplicates. C: static variables, recursion output, pre/post increment. Aptitude: time and work, speed and distance.
Day 5Matrices: spiral print, rotate 90 degrees, search a small matrix inside a big one. C output set 4 and 5. Aptitude: averages, ages, number series.
Day 6Recursion and stack: balanced parentheses, remove unbalanced parentheses, decode "3[a2[bc]]". C: strings and strlen, sizeof traps. Aptitude: probability, permutations.
Day 7Mock round 1. 10 C output plus 10 aptitude, 60 minutes, on paper, no notes. Score it. Redo every wrong C question until you can explain it. Evening: 3 Zoho set-1 problems from the GitHub repo, 3 hours, timed.
Week 2: Sept 29 to Oct 5. Round 2 speed, round 3 design, OOP
Day 8OOP in Python: class, object, __init__, methods, inheritance, encapsulation. Explain the four pillars aloud. Build a tiny Bank Account class with deposit, withdraw, error on overdraft.
Day 9First machine-round project: Railway reservation. Modules: book ticket, check availability, cancel, waiting list, report. 20 minutes of design on paper first (entities, functions), then 3 hours coding. Save it in this repo.
Day 10Refactor day 9 project: split into functions and classes, handle invalid input, add a menu loop. Then a change request: add RAC quota. Time how long the change takes. Aptitude 20 questions.
Day 11Round 2 mock: 5 problems, 3 hours (group anagrams, spiral matrix, string decoder, alternate sort, pattern). C output 15 questions.
Day 12Second project: Call taxi booking. Taxis at points A to F, assign nearest free taxi, fare 100 for first 5 km plus 10 per km, earnings report. Same routine: design, happy path, then edge cases.
Day 13Data structures you must be able to explain and use: list vs linked list, stack, queue, dictionary (hash map), tree basics, BFS. Big-O of each operation. C output 15 questions.
Day 14Full day-1 simulation: round 1 (60 min) then round 2 (3 h). Evening off.
Week 3: Oct 6 to 8. Interview prep and polish
Day 15Resume deep dive. For each project (gastric cancer multi-omics, KRAS G12D design, DuoGrow, RAG assistant, SIH frontend) write: problem, your exact role, tools, one design decision and why, one thing that broke and how you fixed it. Practise saying each in 2 minutes.
Day 16ZLabs basics: what is training vs inference, overfitting, train/test split, what a model file is, what latency and throughput mean, what a GPU does, what data drift is. Link every answer to your RAG assistant and the omics work. SQL: select, where, group by, join, 10 queries.
Day 17, Oct 8Light: 10 C output, 10 aptitude, re-run one project from memory. Pack laptop, charger (battery is failing, sit near a socket), pen, ID, printed resume. Sleep.
ZLabs NLP / ML round
Reported ZLabs format: a seminar or discussion on a given ML/DL topic, plus questions on ML basics. One NLP candidate was asked to label extracted text and train a model to pull out fields. This is the round where your background is an advantage. Prepare to explain each of these in two minutes with one concrete example from your own work:
- Tokenisation, stop words, stemming vs lemmatisation, bag of words, TF-IDF.
- Word embeddings (word2vec idea), then transformers and attention in plain words. Why BERT differs from GPT.
- What fine-tuning is, what a pretrained model file is, inference vs training.
- Your RAG assistant end to end: chunking, embedding, vector search, prompt, answer. Design decisions and what broke.
- Train/validation/test split, overfitting, precision vs recall vs F1, confusion matrix.
- Deployment vocabulary from the job post: latency, throughput, batching, GPU utilisation, data drift, monitoring. Say what each means and how you would measure it.
- Named entity recognition and text classification: how you would build one from labelled data with scikit-learn or Hugging Face.
Fold this into days 15 and 16, 90 minutes each, and be ready to give a 10-minute seminar on "how a transformer works" or "how RAG works" with a diagram on paper.
Rules that come from the failure stories
- Write answers, do not guess. Round 1 has no options. Trace the C code line by line on paper.
- Talk while you code. Evaluators walk around in rounds 2 and 3 and score your explanation. Silence reads as confusion.
- Design first in round 3. 20 minutes on paper: entities, functions, data structures. Then happy path. Then input validation. Then extras.
- Never use a library where a loop is expected.
sorted(), reversed(), str.split() may be disallowed. Know how to do them by hand.
- Know your projects better than your DSA. The candidate who reached the ZLabs technical round and failed was a strong Java developer who could not go deep on ML.
- Solve 2 problems fully rather than 5 half-done. Candidates with 2 clean solutions advanced; partial solutions to all 5 did not.
Sources used
- GfG 2025: 600 to 32 to 5 to 1 funnel, exact round-2 problems
- GfG 2024 on-campus: 1500 to 250, social-media app in round 3, rejected in tech round
- GfG ZLabs ML interview: round-1 problems, failed on ML depth
- 2026 freshers guide: machine round eliminates the most, design-first strategy
- PrepInsta round structure and advanced problems
- Glassdoor ZLabs ML: DSA, advanced DSA, ML/DL seminar, aptitude, HR
- GitHub: past Zoho question sets
- Railway reservation round-3 walkthrough (Tamil), Taxi booking walkthrough (Tamil)
Reddit could not be fetched directly; the failure reasons above come from LeetCode Discuss, GeeksforGeeks and Code360 posts, which say the same things Reddit threads do: round 3 and the technical interview are where people fall.