┌─────────────────────────────────────────────────────────────────────┐
│ 📅 Started: 2025 · 🎯 Target: ₹15–25 LPA AI/ML Engineer Role │
│ 🧠 5 Phases · 33 Weeks · 200+ Daily Study Plans · No Time Limit │
│ 🏗️ Built by a B.Tech CSE (AI & ML) student, for every student │
└─────────────────────────────────────────────────────────────────────┘
Okay real talk —
You know that feeling when you Google "how to learn ML" and you get 47 Medium articles, 12 YouTube playlists, 3 Udemy course ads, and absolutely zero idea what to actually do tomorrow morning?
Yeah. I lived that. I was that guy staring at a roadmap that said "learn Python → learn ML → get job" like bro WHAT DOES THAT MEAN.
So I said forget it and built my own. Day by day. Week by week. Topic by topic.
This is the exact plan I'm following as a 3rd-year B.Tech AI & ML student at KMCE, Hyderabad to go from
Every single day has a topic. Every week ends with something built. Every phase ends with a deployed project. No vague "learn deep learning" nonsense — I mean Monday: build a 2-layer neural net in pure NumPy. Friday: deploy it via FastAPI on Render.
This is what I open every morning. Now it's yours too.
| File | What it is |
|---|---|
ml_day_by_day_plan.html |
The OG plan — Phase 1, 2, 3 with every single day mapped out |
ml_day_by_day_plan_v2.html |
The upgraded version — adds Phase 2.5 (AI Engineer Layer) + Phase 5 (GenAI Stack) |
ml_complete_4phase_plan.html |
Full plan including Deep Learning — CNNs, RNNs, Transformers, the whole thing |
dsa_aiml_roadmap.html |
DSA track built specifically for ML engineering interviews (not generic LC grind) |
eda_revision_plan.html |
My EDA framework — how I approach any new dataset without panicking |
aiml_engineer_roadmap.svg |
The big picture career map — where I am, where I'm going, how I'm getting there |
💡 Quick tip: Download and open the HTML files in your browser — they're interactive dashboards with clickable day-by-day detail. GitHub's preview doesn't do them justice.
PHASE 1 ──────────────────────────────────────────── FOUNDATIONS
Weeks 1–6 │ Python · NumPy · Pandas · Matplotlib · Linear Algebra
│ Probability · Linear Regression · 2 Mini Projects
└─ Capstone: Full EDA on Titanic Dataset
PHASE 2 ──────────────────────────────────────────── CORE ML
Weeks 7–12 │ scikit-learn Pipelines · Logistic Regression · KNN
│ Decision Trees · Random Forest · K-Means · PCA
└─ Capstone: Kaggle House Prices (first public submission)
PHASE 2.5 ────────────────────────────────────────── AI ENGINEER LAYER ✦
Weeks 13–15│ SQL → FastAPI → Claude/OpenAI API + Prompt Engineering
│ Serve your ML model as a live REST API
└─ Capstone: TruthLens News Analyzer — live URL on Render
PHASE 3 ──────────────────────────────────────────── ADVANCED ML
Weeks 16–18│ Feature Engineering · XGBoost · LightGBM
│ NLP Basics · TF-IDF · Sentiment Analysis
└─ Capstone: TruthLens v2 — replace rules with real ML
PHASE 4 ──────────────────────────────────────────── DEEP LEARNING ✦
Weeks 19–27│ NN from Scratch (NumPy) → PyTorch → CNNs → RNNs/LSTMs
│ Transformers · Attention · BERT · Fine-tuning · LoRA
└─ Capstone: BERT-powered Fake News Detector (TruthLens 3.0)
PHASE 5 ──────────────────────────────────────────── GENAI STACK ✦
Weeks 28–33│ LangChain · ChromaDB · RAG Pipelines · AI Agents
│ LangGraph · Docker · AWS · Cloud Deployment
└─ Capstone: Smart News Analyzer — Full AI App, Live on AWS
🐍 Phase 1 — Foundations (Weeks 1–6)
No skipping. No "I already know some Python." Trust the process.
| Week | Python | Mathematics | ML Concept |
|---|---|---|---|
| 1 | Variables, Lists, Dicts | Vectors as lists | What is ML? Supervised vs Unsupervised |
| 2 | Loops, Functions, List Comprehensions | Dot Product | ML Workflow: data → train → predict |
| 3 | NumPy arrays, Broadcasting | Matrices, np.dot() |
Cosine similarity from scratch |
| 4 | Pandas DataFrames, Merge, Groupby | Derivatives (concept) | Linear Regression theory |
| 5 | Matplotlib, Seaborn, EDA patterns | Probability basics | LinearRegression in scikit-learn |
| 6 | MINI PROJECT | — | Full EDA on Titanic → push to GitHub |
The weekly rhythm: Python (Mon) · Math (Tue) · Python (Wed) · Math (Thu) · ML (Fri) · Revise (Sat) · Rest (Sun)
After 6 weeks you can: write Python without Googling basic syntax, manipulate any dataset with NumPy + Pandas, build a linear regression model, and make visualizations that actually look good. Most importantly — you'll have your first GitHub commit streak going and something real pushed online.
🤖 Phase 2 — Core ML (Weeks 7–12)
This is where things click. You stop feeling like a fraud.
| Week | Topic | Key Concept | Hands-on |
|---|---|---|---|
| 7 | Linear Regression deep dive | Cost function, MSE, R² | California Housing dataset |
| 8 | Logistic Regression + KNN | Sigmoid, Precision/Recall/F1 | Iris classification |
| 9 | Decision Trees + Random Forest | Gini impurity, Bagging | Titanic survival >80% |
| 10 | Model Evaluation | Cross-validation, GridSearchCV, Bias-Variance | Tune everything |
| 11 | Unsupervised — K-Means + PCA | Centroids, Dimensionality reduction | Cluster Iris without labels |
| 12 | KAGGLE MILESTONE | Full end-to-end pipeline | House Prices — real public submission |
After Phase 2 you can train, evaluate, tune, and submit a real model to Kaggle. That leaderboard rank on your resume hits different than a course certificate.
⚙️ Phase 2.5 — AI Engineer Layer (Weeks 13–15) ✦
This phase exists because I went through actual Hyderabad job listings (Impressico, Redsage, Janooma) and realized pure ML modelling wasn't enough. Every posting also wanted SQL, API frameworks, and LLM experience. So I added it.
| Week | Skill | What Gets Built |
|---|---|---|
| 13 | SQL — SQLite, JOINs, GROUP BY, Python integration | Import Titanic into SQLite, write 5 analytical queries |
| 14 | FastAPI — REST APIs, Pydantic models, Swagger UI, Render deploy | Your Titanic Random Forest served as a live public API |
| 15 | Claude/OpenAI API + Prompt Engineering | News Analyzer CLI → wrapped in FastAPI |
💬 The interview answer you now have: "I deployed a Random Forest model as a REST API using FastAPI, with JSON input/output, auto-generated Swagger docs, and it's live on Render right now." That's it. That's the answer that gets you the callback.
After Phase 2.5 you can apply to: Redsage Prompt Engineer (Hyderabad), FastAPI + ML Engineer roles, AI developer trainee positions across India.
🚀 Phase 3 — Advanced ML (Weeks 16–18)
You know how to train models. Now you learn to make them actually good.
- Week 16: Feature engineering — encoding, scaling, creating interaction terms, chaining everything into sklearn Pipelines
- Week 17: XGBoost + LightGBM — this is what wins Kaggle. Target top 20% on House Prices.
- Week 18: NLP basics — NLTK, TF-IDF, sentiment analysis. Also: TruthLens gets upgraded from a heuristic rules engine to a real trained classifier. Big glow-up.
🧠 Phase 4 — Deep Learning (Weeks 19–27) ✦
The part everyone hyped up but nobody explained properly. We go bottom-up.
| Week | Topic | What You Build |
|---|---|---|
| 19 | Neural Networks from NumPy scratch | XOR classifier — zero libraries, pure math |
| 20 | PyTorch fundamentals | Reimplement Week 19 in PyTorch (you'll cry at how short it is) |
| 21 | CNNs — Conv2d, MaxPool, feature maps | MNIST digit classifier targeting >98% accuracy |
| 22 | RNNs + LSTMs — sequences, hidden state, gates | IMDb sentiment analysis with LSTM |
| 23 | Transformers — Self-attention, Q/K/V matrices | Run BERT sentiment in literally 10 lines via HuggingFace |
| 24 | BERT architecture + fine-tuning | Custom text classifier with HuggingFace Trainer API |
| 25 | LoRA / QLoRA — efficient fine-tuning | distilBERT-powered TruthLens (your best project so far) |
| 26–27 | CAPSTONE | BERT Fake News Detector — deployed, shareable, portfolio-ready |
🌐 Phase 5 — GenAI Stack (Weeks 28–33) ✦
This is where ₹8 LPA becomes ₹20 LPA. The GenAI layer is what companies are actually hiring for right now.
| Week | Topic | What Gets Built |
|---|---|---|
| 28 | LangChain — chains, memory, document loaders | PDF Q&A chatbot |
| 29 | ChromaDB + RAG pipelines | "Ask My Notes" — upload your study PDFs, ask anything |
| 30 | AI Agents + LangGraph | Agent that uses web search + your ML model + calculator as tools |
| 31 | Docker + AWS/Render deployment | Everything goes live at a real public URL |
| 32–33 | FINAL CAPSTONE | Smart News Analyzer — ML + FastAPI + SQLite + Claude API + RAG + AWS |
Roles this directly unlocks: LLM Systems Engineer at Janooma (Bengaluru, ₹8–12 LPA), GenAI Python Full Stack at Impressico (Hyderabad), Prompt Engineer at Redsage (Hyderabad).
Not certificates. Not course completions. Actual things deployed on the internet with real URLs.
| # | Project | Stack | Phase |
|---|---|---|---|
| 1 | Titanic EDA Notebook | Pandas, Seaborn, Matplotlib | Phase 1 |
| 2 | Iris Classifier | scikit-learn, LogReg, KNN | Phase 2 |
| 3 | Kaggle House Prices | Random Forest, XGBoost, GridSearchCV | Phase 2 |
| 4 | TruthLens v1 | Rules engine, NewsAPI, Streamlit | Phase 2.5 |
| 5 | Survival Predictor API | FastAPI, joblib, deployed on Render | Phase 2.5 |
| 6 | News Analyzer | Claude API, Prompt Engineering, FastAPI | Phase 2.5 |
| 7 | TruthLens v2 | TF-IDF + Logistic Regression (real ML now) | Phase 3 |
| 8 | MNIST CNN Classifier | PyTorch, Conv2d, >98% accuracy | Phase 4 |
| 9 | IMDb Sentiment (LSTM) | PyTorch, LSTM | Phase 4 |
| 10 | TruthLens 3.0 | distilBERT, LoRA, HuggingFace Trainer | Phase 4 |
| 11 | "Ask My Notes" RAG App | LangChain, ChromaDB, FastAPI | Phase 5 |
| 12 | Smart News Analyzer | Full stack: ML + FastAPI + SQLite + Claude + RAG + AWS | Phase 5 |
The ML roadmap alone won't get you past technical interviews at the companies that pay well. So there's a separate DSA track running alongside.
Tier 1 Arrays · Strings · HashMaps · Two Pointers · Sliding Window
Tier 2 Recursion · Sorting · Binary Search · Stack · Queue
Tier 3 Trees · Graphs (BFS/DFS) · Heaps · Dynamic Programming
Tier 4 Tries · Advanced Graphs · System Design Fundamentals
Check dsa_aiml_roadmap.html for the full problem list with LeetCode tags and study order. The target is LeetCode Medium comfort level — that's the bar for product company ML roles.
The schedule: 3–4 problems/week while deep in ML learning, ramp to 5–7/week during interview prep mode.
Every single dataset in this roadmap goes through the same framework before any model gets trained. It's in eda_revision_plan.html with full examples.
I — Inspect Shape, dtypes, head/tail, basic sanity checks
Q — Quality Nulls, duplicates, outliers, weird values
V — Visualize Distributions, correlations, pairplots, heatmaps
C — Correlate Feature-feature and feature-target relationships
S — Summarize Written findings — what does this data actually say?
The S step is what most people skip and it's exactly what interviewers ask about. "What did you find in the EDA?" You need an actual answer, not "I ran .describe()."
You'll get the most out of this if you're:
- A B.Tech CSE / AI / ML student who wants a job that actually pays, not just a degree to frame
- A CS grad who spent 4 years in theory and came out not knowing how to train a model
- Someone who's been watching ML tutorials for 6 months and hasn't shipped anything yet
- Anyone targeting AI/ML roles in Hyderabad, Bengaluru, or remote — and wants to know exactly what to learn next
- The kind of person who would rather build something broken on Friday than watch a perfect tutorial on Sunday
This is probably not for you if:
- You want quick wins and certificates you can screenshot for LinkedIn (no shade, just not what this is)
- You're looking for academic-depth research material — this is engineering, not academia
Updated as I go
| Phase | Status | Notes |
|---|---|---|
| Phase 1 — Foundations | ✅ Done | Titanic EDA notebook pushed to GitHub |
| Phase 2 — Core ML | 🔄 In Progress | Currently on model evaluation + sklearn |
| Phase 2.5 — AI Engineer Layer | ⏳ Up Next | — |
| Phase 3 — Advanced ML | ⏳ Upcoming | — |
| Phase 4 — Deep Learning | ⏳ Upcoming | — |
| Phase 5 — GenAI Stack | ⏳ Upcoming | — |
Daily commits at Sampath7890/ML-phase1-foundation — green squares every weekday, no excuses.
Same structure every week. No decision fatigue about what to do today.
Monday ── New concept (Python / ML / Deep Learning)
Tuesday ── The math behind Monday's concept
Wednesday ── Code it from scratch
Thursday ── Second technique or deeper math
Friday ── Build something with it (this is non-negotiable)
Saturday ── Write 5 Q&A from the week in a revision notebook
Sunday ── Rest. Seriously. Rest.
One rule I follow: Git commit every weekday. Even if it's 5 lines of code and a comment. The streak is the discipline — it forces you to sit down even when you don't feel like it.
- 3Blue1Brown — Essence of Linear Algebra — watch this before anything else, seriously
- StatQuest with Josh Starmer — he explains everything like you're a person, not a PhD student
- 3Blue1Brown — Neural Networks — backpropagation finally makes sense
- Krish Naik's ML playlist — very implementation-focused, great for the coding side
- Fast.ai Practical Deep Learning — top-down approach, you build first then understand why
- HuggingFace NLP Course — the best free resource for transformers and fine-tuning
- Andrej Karpathy — Neural Networks: Zero to Hero — you build GPT from scratch, unreal content
- Kaggle Learn — free micro-courses, surprisingly good for feature engineering and SQL
- Striver's A2Z DSA Course — the most structured DSA path I've found
- NeetCode Roadmap — pattern-based, great for recognizing problem types fast
- LeetCode Medium — the minimum bar for product company interviews, no way around it
- FastAPI Official Docs — genuinely one of the best-written docs in open source
- Anthropic API Docs — Claude API for building with LLMs
- LangChain Docs — RAG and agent workflows
- HuggingFace Transformers Docs — your home base for Phase 4 and 5
If you find a better resource, want to add a notebook, or spot something wrong — PRs are very welcome. This roadmap is a living thing.
# Fork → branch → change → PR
git checkout -b feature/what-you-changed
git commit -m "Add: brief description"
git push origin feature/what-you-changedThings that would genuinely help:
- Jupyter notebooks for any week (runnable on Google Colab)
- Week-by-week quiz questions
- Hindi/Telugu translations of the roadmap structure
- "I followed this and here's what happened" stories in Discussions — love reading those
- Jupyter notebooks for Phase 1 and 2 — fully runnable on Colab, no setup needed
- Auto-updating progress tracker via GitHub Actions
- Phase 5 (GenAI Stack) full day-by-day expansion
- "Mistakes I made" section — the dumb errors at each phase so you can skip them
- MLOps track — MLflow, DVC, model monitoring (post-Phase 5)
- Data Engineering branch — Airflow, Spark, dbt
- ML System Design prep — 50 questions with answers
- Discord community for people following this roadmap (DM me if you want in early)
Alright here's the thing.
I'm a 3rd-year B.Tech AI & ML student in Hyderabad. When I started seriously trying to learn AI/ML, every roadmap I found was either "here are 200 resources, figure it out" or some bootcamp trying to sell me a ₹50k course.
Nobody told me what to do on a Tuesday at 9am. Not what topics exist. What to literally open and type.
So I built that for myself. Iterated on it for months. Added Phase 2.5 when I realized that ML theory alone wasn't enough for the job listings I was reading. Added Phase 5 when I saw where the actual hiring was moving. Rebuilt the whole thing when I found better ways to structure the progression.
It's not perfect. I'm still following it. But it's mine — and now it's yours too.
If this saves even one person from the 3-month YouTube rabbit hole I went through — worth it.
And if it helps you land that AI/ML role — drop a ⭐ and let me know. Genuinely.
G Sampath Kumar — B.Tech CSE (AI & ML), KMCE Hyderabad, Batch of 2028
Building in public. Daily commits. No time limit.
If this helped you at all — ⭐ star it. It literally takes 2 seconds and it means a lot.
"Everyone was a beginner once. Some of them just committed to GitHub every day."