College Engineering AI Projects India 2026: Mini-Project to Capstone Playbook

Engineering projects in India have become the single biggest differentiator between a student who lands a product-company offer and one stuck in mass-recruitment pools. As placement data from 2025-26 makes clear, companies now weigh a strong GitHub portfolio over college tier for AI roles. This playbook is the end-to-end workflow for Indian BTech, BE, and Integrated MTech students to go from a semester-5 mini-project to a capstone that actually gets interview calls.

Every framework, topic, and mentor route below is free. Every project pattern is AICTE-compliant, fits the typical 2-semester capstone window, and uses the 2026 tech stack that Indian product companies (Flipkart, Razorpay, Zomato, Cred, PhonePe, Swiggy, Jio Platforms) are actually hiring for.

What You Will Learn

• What AICTE expects from mini-projects and capstones
• The 2026 AI tech stack you should actually learn
• 10 battle-tested project ideas (mini + capstone tier)
• How to pick a topic your guide will approve in one meeting
• How to ship to GitHub like a professional
• How to find external mentors
• What recruiters look for during placements

AICTE Mini-Project vs Capstone — What Is Required

Under the AICTE Approval Process Handbook 2025-26, BTech and BE programmes are expected to include:

• Mini-project — typically semester 5 or 6, 2 to 4 credits, team of 2-4, 6-8 weeks, demonstrating application of a known technique on a new problem or dataset.
• Capstone / final-year project — semester 7 and 8 combined, 8-12 credits, team of 2-4, full academic year, requiring novel integration, testing, formal report, and viva.

Individual universities (Anna, VTU, Mumbai University, JNTU, SPPU) publish rubrics with specific word counts, format, and plagiarism limits. Check your own university's project circular for the year. AICTE also now encourages Work Integrated Learning Programs (WILP) where the project tackles a real company problem.

The 2026 AI Tech Stack for College Projects

Pick one item from each layer. Do not try to learn everything.

| Layer | Productive 2026 Pick | Why | |-------|---------------------|-----| | Language | Python 3.12 | Default for all AI work | | Deep Learning | PyTorch 2.5 | Industry standard; TensorFlow 2.16 also fine | | NLP / LLM | Hugging Face Transformers | Model zoo + inference + training | | LLM orchestration | LangChain 0.3 or LlamaIndex | Build RAG and agent pipelines | | Experiment tracking | MLflow | The most widely adopted open-source MLOps tool | | Vector DB | Chroma or Qdrant | Free, embeddable | | Backend | FastAPI | Cleanest Python API framework | | Frontend | Streamlit (quick) or Next.js 15 (polished) | Streamlit for viva; Next.js for portfolio | | Containerisation | Docker | Required for reproducibility | | Compute | Google Colab (free) or Kaggle kernels | Enough for 70% of capstones |

For deeper dives, read our RAG for beginners guide and the AI agents tutorial 2026.

10 Project Ideas — Mini and Capstone Tier

Mini-Project Tier (6-8 Weeks)

1. Resume ranking API using BERT embeddings — paste a JD, rank 10 resumes. Fit: CSE, IT. Stack: Hugging Face + FastAPI.
2. Hindi-English medical chatbot for ASHA workers — RAG on translated WHO guidelines. Fit: AI/ML specialisation. Stack: LangChain + Chroma + GPT-4o-mini or Gemini free API.
3. Indian crop-disease detector mobile-cam — fine-tune MobileNetV3 on PlantVillage + iCAR Indian crop dataset. Fit: ECE with embedded. Stack: PyTorch + TFLite.
4. Kannada/Tamil voice assistant for elder care — Whisper for ASR + gTTS for speech. Fit: Any branch. Stack: Hugging Face + Streamlit.
5. Traffic-signal violation detector — YOLOv9 on Bengaluru CCTV sample. Fit: ECE, EEE. Stack: Ultralytics + OpenCV.

Capstone Tier (2 Semesters)

6. Multilingual RAG for state government schemes — ingest MyScheme.gov.in PDFs, answer in 5 Indian languages. Novelty: language routing + citation.
7. Real-time lecture note generator using Whisper + LLM summariser — plug into your own college Zoom. Novelty: on-device inference for privacy.
8. Autonomous campus delivery bot (simulation) — ROS 2 Humble + reinforcement learning in Gazebo. Fit: Mechanical, ECE.
9. GAN-based saree pattern generator with fashion-designer UX — diffusion model fine-tuned on your own scraped dataset of 5,000 sarees (public Indian e-commerce). Novelty: cultural dataset + designer workflow.
10. MLOps pipeline for an Indian fintech credit-risk model — MLflow + Kubeflow + drift detection + Docker + GitHub Actions. Novelty: end-to-end deployment, strongest capstone option for placement-oriented students.

As a 2026 industry analysis put it, capstones that use both modern LLMs and older AI frameworks — for example, RAG combined with computer vision for real-time tracking — are what employers value most.

How to Pick a Topic Your Guide Will Approve in One Meeting

1. Start from a real Indian problem, not from a paper. Saree-pattern generation beats MNIST digit classification at the viva and at the interview.
2. Bring three options to the first meeting — guides love choice.
3. Include a one-page scope doc — problem, data source, expected deliverable, evaluation metric, 8-week timeline.
4. Keep it deployable — a demo URL that loads during viva wins marks.
5. Avoid "build an LLM from scratch" — your guide will politely reject it. Use existing models + your wrapper.

The Data-Access Cheat Sheet

Data is the bottleneck for most Indian capstones. Here are free, legal sources used by top projects:

• data.gov.in — Government datasets (agriculture, health, transport)
• Kaggle India competitions — labeled, ready to use
• AI4Bharat — Indian language datasets (IIT Madras-led)
• RBI Handbook of Statistics — finance
• WHO, ICAR, IMD open APIs — health, agri, weather
• Your own college — ask registrar for anonymised academic data

Avoid scraping copyrighted platforms. Facebook, Instagram, and paid OTT feeds will get your project rejected for ethics.

Shipping to GitHub Like a Professional

A GitHub repo that gets you hired at Flipkart or Razorpay has six things:

1. Clear repo name — multilingual-scheme-rag beats final-year-project
2. README with demo GIF — record with peek or Kap, compress to under 5 MB
3. Architecture diagram — excalidraw.com is free and dev-style
4. Commit hygiene — 30+ meaningful commits, not one "initial commit"
5. License — MIT for portfolio; Apache 2.0 if you fear patent issues
6. Releases tag — v0.1 after mini-viva, v1.0 at capstone defence

Follow the AI portfolio India guide for the exact README template used by hired candidates.

Anti-Patterns That Kill a Portfolio

• Untitled.ipynb in the repo root
• Commit message "final changes again"
• No license file
• Screenshots that show the project crashing
• Empty requirements.txt

Finding a Mentor Outside Your College

Your college guide is usually overloaded with 10-12 teams. To stand out, add an external mentor.

1. Open-source programs — Google Summer of Code, GirlScript Summer of Code, Outreachy, LFX Mentorship. Indian students routinely get in.
2. IIT Bombay's Eklavya programme — takes final-year external students during the summer.
3. LinkedIn cold outreach — write a 4-line message to AI researchers at Microsoft Research India, Google Research Bangalore, Adobe Research Noida, Flipkart AI, or Walmart Labs India. Ask for 20 minutes, not a mentorship. If the call goes well, the mentorship follows.
4. Prof. discovery via arxiv-sanity — find a paper that excites you, DM the lead author. Indian profs respond to specific, respectful asks.

What Recruiters Actually Check During Placements

Product-company recruiters typically spend under 90 seconds on your project review. They check:

• Is the README demo convincing in 30 seconds?
• Does the architecture diagram look production-grade?
• Is the code clean (Black/Ruff formatted, typed)?
• Are there tests (even 3 pytest tests beat zero)?
• Did the student solve a real problem, or just replicate a tutorial?
• Did they measure impact (latency, accuracy, cost)?

If all six are yes, you get a Round 1 call. Read companies hiring AI in India for role-specific expectations.

Timeline — A Real 2-Semester Capstone Plan

| Month | Deliverable | |-------|-------------| | July | Topic shortlist + scope doc | | August | Literature review (10 papers), baseline reproduction | | September | Data pipeline + MVP v0.1 | | October | Feature complete v0.5 + mid-viva | | November | Evaluation + paper draft for a student conference | | December | Deployment + Docker build | | January | Freeze + regression testing | | February | Report writing (40-60 pages) | | March | Final viva + open-source release |

Publishing a short paper at IEEE student conferences (SCONEST, ICETET, COMSNETS Student Track) or a workshop at ICMLA dramatically improves your MS application chances.

Key Takeaways

• AICTE capstones are worth 8-12 credits — treat them as a product, not an assignment
• The 2026 default stack: PyTorch + Hugging Face + LangChain + MLflow + FastAPI + Docker
• Pick a real Indian problem; free data is available via data.gov.in and AI4Bharat
• A strong GitHub README is a 90-second interview
• External mentors are accessible via GSoC, Eklavya, and targeted LinkedIn outreach
• Next read: AI portfolio India, the AI internship India guide, and AI career path India

Last updated: April 19, 2026