25 Project Ideas That Actually Impress Recruiters (Branch-Wise for 2026)
Your College Project Is Not Impressing Anyone. Let Us Fix That.#
I need to be honest with you about something. That "Library Management System" on your resume? The "Student Attendance Portal"? The "Online Shopping Website"? Every syllabus asks for them, so every resume has them. A recruiter who has seen the same title fifty times learns nothing about you from it. That is our opinion, not a survey result.
Here is the thing though. You do not need to build the next Swiggy or CRED to stand out. You just need to build something that shows three qualities:
- You can solve a real problem (not a textbook exercise)
- You can use modern, relevant tools (not just what your college syllabus covers)
- You can explain what you built and why it matters (the resume bullet is as important as the project)
This guide gives you 25 project ideas across 5 engineering branches. Each one covers what to build, which tech stack to use, what the project shows, how long it takes, and how to write it on your resume. Most of them are 2-4 weeks of part-time work if you already know the basics, and every one can be built with free tools. Free tiers change often, so each project says what we checked.
No more Library Management Systems. Let us build something that actually gets you shortlisted.
Get the next guide by email
New guides and fresh jobs in one short email. Free.
What We Re-Checked in November 2026#
We first published this list in February 2026. In November 2026 we went through every project and checked the tools, libraries, datasets and free tiers against their official docs. A lot had moved. Here is what changed:
- Scraping X (Twitter) for data is out. Third-party pricing guides report the free X API tier closed to new developers in February 2026 (we could not confirm this on X's own site). The hate speech project now uses published research datasets.
- MediaPipe Hands is the old API. Google's MediaPipe guide lists Hands as upgraded to hand landmark detection. The gesture robot now uses the newer one.
- PyPDF2 is finished. Its PyPI page says 3.0.1 (December 2022) is the last version and development moved to pypdf.
- SendGrid and Railway are no longer free forever. Twilio retired SendGrid's free plans in May 2025, and Railway now gives a one-time $5 trial. The job alert bot uses Telegram and a scheduled GitHub workflow instead.
- Amazon's old product API is gone. Amazon says PA-API 5.0 is deprecated and replaced by the Creators API. The price tracker was rewritten around that.
- Crunchbase's free API is closed to new users, per Crunchbase's own docs. We also could not confirm the Tracxn "free reports" mentioned in February, so that link is gone.
- Koo shut down in July 2024, so it is no longer listed as an employer for the hate speech project.
- Several February claims had no source (recruiter quotes, "high demand" lines, a long list of company names, sample accuracy numbers). We removed them or replaced them with JobRise job data.
What the Live Job Data Says (10 Nov 2026)#
Method: JobRise job data, 10 Nov 2026. Our job database is built from public job boards and hiring systems. On that day it held 31,428 open roles in India after removing duplicate postings. The skill counts below are roles our system tagged with that skill from the job text. The word counts at the end of this section only look at the job title and company name.
| Skill | India roles tagged |
|---|---|
| Python | 2,674 |
| SQL | 1,937 |
| Machine learning | 1,614 |
| Kubernetes | 1,219 |
| Large language models | 1,139 |
| Microservices | 922 |
| Docker | 904 |
| React | 734 |
| PostgreSQL | 587 |
| Apache Kafka | 473 |
| Node.js | 443 |
| Statistics | 338 |
| Tableau | 293 |
Inside our data and engineering categories the numbers get smaller but still show where the demand is. In the 2,728 India data roles: NLP 129, PyTorch 121, scikit-learn 78, computer vision 47, FastAPI 43, Hugging Face 36. In the 2,696 India engineering roles: Firmware 69, Embedded systems 40.
Counting only job titles and company names: 80 India roles say "embedded", 58 say "firmware", 20 say "FPGA", 18 say "IoT" and 8 say "CFD".
Read this part carefully. Our crawlers cover company career systems and job boards, which lean towards software and data roles. Campus drives, core mechanical hiring and many ECE roles are under-covered. Small counts for FPGA or CFD do not mean nobody hires for them. They mean we cannot see much of it. The big counts for Python, SQL and cloud tools are real, though, and they are why the CSE and data projects below use that stack.
Before You Start: The 3 Rules of Resume Projects#
Rule 1: The project must solve a problem you can explain in one sentence. If you cannot describe what your project does and why it matters to a non-technical person, it is not a good resume project. "I built a weather app" is weak. "I built a tool that sends automated weather alerts to farmers in drought-prone districts" is strong. Same complexity, completely different impression.
Rule 2: Deploy it or it does not count. A project that only runs on localhost is a learning exercise, not a portfolio piece. Put your code on GitHub with a proper README. If it is a mobile app, record a demo video. Hosting options we checked on 10 Nov 2026:
- Vercel's Hobby plan is free but limited to non-commercial, personal use. Fine for a portfolio.
- Render's free tier gives 750 free instance hours a month. A free web service spins down after 15 minutes without traffic, so the first visit after a break is slow. Free Postgres expires 30 days after creation, so do not keep data you care about there.
- Railway gives a one-time $5 trial credit for up to 30 days, then a Free plan with $1 of credit per month. Good for a short demo, not for something that must stay up.
Rule 3: Write the resume bullet before you start building. Write the bullet you want, then build the project to match it. The bullets below use brackets like [N] for results. Fill them with numbers you actually measured. Never copy a number from a blog post, including this one.
Here is the format for every resume bullet: "Built [what] using [tech stack] that [specific outcome/metric]."
The "Why recruiters like it" lines below are our judgment of what each project proves. They are not recruiter survey results.
CSE / IT Projects (8 Ideas)#
1. Real-Time Code Collaboration Tool
What to build: A browser-based code editor where multiple people can write code simultaneously (like Google Docs, but for code). Include syntax highlighting, cursor tracking, and a basic chat sidebar.
Tech stack: React or Next.js (frontend), Node.js with Socket.IO 4.x (it uses WebSocket and falls back to HTTP long-polling when needed), Monaco Editor (the code editor from VS Code), and Yjs for conflict-free syncing (it is a CRDT framework and lists Monaco as a supported editor binding). Redis is optional, for session state. Redis 8 is available under AGPLv3 again, and Valkey is the BSD-licensed fork if you want the simplest licence.
Why recruiters like it: It shows real-time systems work and that you understand why two people typing at once is a hard problem. Using Yjs gives you an honest way to talk about CRDTs in an interview without claiming you invented one.
Time to complete: 3-4 weeks.
Resume bullet: "Developed a real-time collaborative code editor supporting [N] simultaneous users with [X] ms median sync latency (measured in testing), using React, Socket.IO, Yjs and Monaco Editor, deployed on Vercel"
2. AI-Powered Resume Analyzer API
What to build: A REST API that accepts a resume (PDF) and a job description (text), extracts key information using NLP, and returns a compatibility score with specific improvement suggestions.
Tech stack: Python, FastAPI, spaCy or Hugging Face Transformers for NLP (the current Transformers README asks for Python 3.11+ and PyTorch 2.6+), pypdf for PDF parsing (not PyPDF2, which is no longer developed), deployed on Render's free tier.
Why recruiters like it: It shows you can work with NLP, build an API, handle file processing, and solve a real problem. You can also use it on your own resume.
Heads up: Use your own resume and fake ones for testing. Real resumes hold phone numbers and addresses, and you should not store other people's.
Time to complete: 2-3 weeks.
Resume bullet: "Built a resume analysis API using FastAPI, pypdf and spaCy that scores resume-JD fit across [N] parameters, checked against [N] manually reviewed resumes with [X]% agreement"
3. Expense Splitting App with UPI Deep Links
What to build: A mobile-friendly web app where groups can log shared expenses (dinner, trips, rent) and settle debts. Generate UPI payment deep links so users can open their UPI app with the amount filled in.
Tech stack: Next.js or React Native, Firebase (Firestore plus Google sign-in), UPI deep links in the upi://pay format, Tailwind CSS.
What we checked: Google's in-app payments doc lists the link parameters pa, pn, mc, tr, tn, am and cu, and says NPCI requires apps to support the generic UPI intent. It also tells merchants to confirm payment with their payment provider before trusting a result. For your app that means: a link opening is not proof that money moved, so add a "mark as settled" button. We did not find which parameters NPCI marks as mandatory, so test your links on two or three UPI apps. On Firebase, the no-cost Spark plan gives 1 GiB of Firestore storage, 50K reads and 20K writes a day, and 50K monthly active users for sign-in. Phone OTP sign-in is not on Spark, so use Google sign-in.
Why recruiters like it: It is India-specific, solves a daily problem, and shows payment-flow thinking and mobile UX.
Time to complete: 2-3 weeks.
Resume bullet: "Created a group expense splitting app with UPI deep link settlement using Next.js and Firebase, tested by [N] users across [N] groups, with automated debt simplification so each group needs fewer payments"
4. GitHub Profile Analytics Dashboard
What to build: A web app where users enter a GitHub username and get a visual dashboard of their coding activity: language distribution, commit patterns, most active repositories, and PR statistics.
Tech stack: Next.js, GitHub REST API, Chart.js or Recharts for visualizations, Tailwind CSS.
What we checked: GitHub's rate limit docs say 60 requests an hour without authentication and 5,000 an hour for an authenticated user. A dashboard that fetches live for every visitor hits 60 fast. Cache results on your server and keep any token on the server, never in browser code.
Why recruiters like it: Clean API integration, data visualization, and a tool developers actually want. The caching and rate limit handling is the part worth talking about in an interview.
Time to complete: 2 weeks.
Resume bullet: "Built a GitHub analytics dashboard visualizing [N] metrics (language distribution, commit patterns, PR stats) for any public profile using the GitHub API and Recharts, with server-side caching to stay under API rate limits"
5. Smart Notification Aggregator Chrome Extension
What to build: A Chrome extension that pulls notifications from a few services into one feed with priority sorting and snooze.
Tech stack: Chrome extension on Manifest V3, React for the popup UI, Chrome Storage API, OAuth2 where a service supports it.
What we checked: Manifest V2 is dead. Per Chrome's MV2 timeline, Chrome 139 and later do not run MV2 extensions, and the Web Store removed all remaining MV2 extensions by 31 Aug 2026. Old tutorials with manifest_version: 2 will not work.
Pick sources that have a documented API. GitHub does: GET /notifications only works with a classic personal access token with the notifications or repo scope. We searched LinkedIn's Marketing API docs and found no documented endpoint that returns a member's personal notifications, so the February idea of aggregating LinkedIn is dropped. Check the docs of every service (Gmail, Slack, anything else) before you promise it in your README.
Why recruiters like it: Permissions, OAuth and background workers are good interview material.
Time to complete: 3 weeks.
Resume bullet: "Developed a Manifest V3 Chrome extension aggregating notifications from [N] services into a unified priority-sorted feed, with snooze and per-source filters"
6. Automated Job Alert Bot
What to build: A bot that checks job listings from companies you choose, filters them by your keywords, role level and location, and sends you new matches on Telegram.
Tech stack: Python, the public job board APIs of Greenhouse and Lever, the Telegram Bot API for alerts, and a scheduled GitHub Actions workflow to run it.
What we checked:
- Greenhouse's Job Board API says "authentication is not required for any GET endpoints". The list call is
GET https://boards-api.greenhouse.io/v1/boards/BOARD_TOKEN/jobs. - Lever's Postings API lists published postings at
api.lever.co/v0/postings/SITE, and says published postings are publicly viewable. - GitHub Actions scheduled workflows run at most every 5 minutes, use UTC, and are switched off in a public repo after 60 days without repository activity. Push a small commit now and then.
- The February version scraped Naukri and Internshala. Naukri's terms of use list crawling or scraping its content as a prohibited use, so we dropped it. We did not confirm Internshala's terms either way, so read the terms and robots.txt of any site before you scrape it.
- SendGrid is gone from the stack. Its free plan was replaced by a 60-day trial. Telegram needs no email service.
Why recruiters like it: It shows API use, scheduling, deduplication and notifications, which are everyday backend skills. You also get to use it for your own job hunt.
Time to complete: 2 weeks.
Resume bullet: "Built an automated job alert bot polling public Greenhouse and Lever APIs for [N] companies on a schedule, filtering by [N] criteria, and delivering deduplicated alerts via Telegram to [N] users"
7. Peer-to-Peer File Sharing with WebRTC
What to build: A browser-based file sharing tool where two users share files directly, without the file passing through your server. Include a simple drag-and-drop UI.
Tech stack: React, the WebRTC data channel API, a small signaling server (Node.js plus Socket.IO) to introduce the two browsers, Tailwind CSS.
What we checked: MDN's data channels guide says most browsers handle messages of at least 256 KB, assumes 64 KB if the other side does not say, and warns that big messages block other traffic. So send files in chunks. It also says channel data is encrypted with DTLS and does not pass through your server. Test on two different networks, because two laptops on your hostel WiFi proves very little.
Why recruiters like it: Peer-to-peer connections, chunking and progress tracking are good interview material.
Time to complete: 2-3 weeks.
Resume bullet: "Created a browser-based P2P file sharing tool using WebRTC data channels with chunked transfers, tested up to [X] MB files across [N] network setups, with a signaling server handling [N] concurrent sessions"
8. Microservices-Based URL Shortener
What to build: A URL shortener built as separate services: URL creation, redirect handling, analytics (click tracking), and user authentication, behind an API gateway.
Tech stack: Node.js with Express 5 (needs Node 18 or higher), Redis or Valkey for caching, PostgreSQL for storage, Docker, Nginx as the gateway.
Honest take: Four services for a URL shortener is more than the problem needs. That is the point of the project, but be ready for an interviewer to ask why. A good answer is "to learn how services talk to each other, and here is what I would not split in production."
Why recruiters like it: It shows architecture thinking, Docker and service-to-service communication. In our data, Docker is tagged on 904 India roles, Kubernetes on 1,219 and microservices on 922 (JobRise job data, 10 Nov 2026).
Time to complete: 3-4 weeks.
Resume bullet: "Architected a microservices URL shortener with [N] independent services (creation, redirect, analytics, auth), containerized with Docker, sustaining [X] redirects/second in load testing on [your hardware] with Redis caching"
Advertisement
ECE / Electronics Projects (5 Ideas)#
9. IoT-Based Smart Energy Monitor
What to build: A device that monitors household energy use in real time and shows it on a web dashboard. Alert users when consumption crosses a threshold.
Tech stack: ESP32 microcontroller, a clamp-on CT (current transformer) sensor, Arduino IDE for firmware, MQTT, a Node.js backend, a React dashboard on Vercel.
What we checked: HiveMQ still runs a public broker, but its own page says it "is public and shared, so it is not intended for private or production data". Anyone can read your topics, and your home energy use is private data. Use HiveMQ's free private broker (the same page says you can create one in under 5 minutes) or run a broker on your own laptop. Vercel's Hobby plan is non-commercial only.
Safety: Use a clamp-on sensor that goes around a wire. Do not open or splice mains wiring without a qualified person with you.
Why recruiters like it: It joins embedded firmware, a messaging protocol and a web dashboard in one project. In our data, India engineering roles tag Firmware 69 times and Embedded systems 40 times, and 18 India job titles mention IoT (JobRise job data, 10 Nov 2026).
Time to complete: 3-4 weeks (including hardware setup).
Resume bullet: "Designed an IoT energy monitor using ESP32 and a CT sensor, publishing readings over MQTT to a React dashboard, with threshold alerts and [X]% error against a reference meter"
10. Gesture-Controlled Robot with Computer Vision
What to build: A small robot controlled by hand gestures seen through a webcam. Recognize 5-6 gestures (forward, back, left, right, stop, speed up) and send commands wirelessly to the robot.
Tech stack: Python, OpenCV, MediaPipe Tasks (Hand Landmarker or Gesture Recognizer), Arduino for the robot, Bluetooth or WiFi for communication.
What we checked: The old mp.solutions.hands code from older tutorials is the legacy API. Google's MediaPipe guide marks Hands as upgraded to hand landmark detection. The Gesture Recognizer installs with pip install mediapipe and ships with these categories: None, Closed_Fist, Open_Palm, Pointing_Up, Thumb_Down, Thumb_Up, Victory, ILoveYou. You could map Open_Palm to stop and Pointing_Up to forward, then read left and right from the hand landmarks. Or train your own gestures.
Why recruiters like it: It combines vision, embedded control and wireless communication, and you can demo it live in an interview.
Time to complete: 3-4 weeks.
Resume bullet: "Built a gesture-controlled robot using OpenCV and MediaPipe hand tracking, recognizing [N] gestures with [X]% accuracy over [N] test attempts, controlling motors wirelessly via Bluetooth"
11. FPGA-Based Digital Signal Processing Pipeline
What to build: A real-time audio signal processing pipeline on an FPGA. Include filters (low-pass, high-pass, band-pass) and an FFT. Take audio in and send processed audio out.
Tech stack: AMD Vivado ML Standard Edition or Quartus Prime Lite, Verilog or VHDL, and a development board your lab has.
What we checked: AMD's Vivado device list (UG973) puts the Artix-7 XC7A35T (the chip on Digilent's Basys 3) in the free Standard Edition. Quartus Prime Lite is free and its device list includes MAX 10 devices, the family on the DE10-Lite, but check the list for your exact chip. Intel's FPGA tools now sit under the Altera name, so search docs.altera.com. Check that your board has audio in and out before you plan around a microphone and speaker. Many do not, and you may need an add-on module.
Why recruiters like it: Most ECE students touch an FPGA only in lab sessions, so a project of your own shows initiative. Be realistic about demand: only 20 India job titles in our data mention FPGA (JobRise job data, 10 Nov 2026), and our coverage of semiconductor hiring is thin. If chip design is your goal, ask seniors and check company career pages directly.
Time to complete: 4 weeks.
Resume bullet: "Implemented a real-time audio DSP pipeline on an Artix-7 FPGA with [N] configurable filters and FFT output, processing audio at [X] kHz sample rate with [X] ms measured latency"
12. LoRa-Based Campus Communication Network
What to build: A long-range, low-power network for a campus using LoRa modules. Send text messages, sensor data or alerts between nodes without WiFi or cellular.
Tech stack: LoRa modules (SX1276/SX1278 family), Arduino or ESP32, a simple web interface for message visualization.
What we checked: Buy modules for the 865 to 867 MHz band. Per DoT, that band is licence-exempt in India under a 2008 notification (GSR 564(E)), and a DoT clarification repeats the limits of 1 W transmitter power, 4 W ERP and 200 kHz bandwidth. LoRaWAN deployments in India use the IN865 plan. Do not assume the 433 or 915 MHz modules you see on shopping sites are legal to use here. We did not verify the later 2021 rules that extend the framework to 868 MHz, so check the DoT site before using it.
Why recruiters like it: It shows RF basics, power management and the limits of a real wireless link, which is different from calling a cloud API.
Time to complete: 3 weeks.
Resume bullet: "Developed a LoRa network on the 865 MHz band connecting [N] nodes for messaging and sensor relay, with measured range of [X] m on campus and [X] hours of battery life per charge"
13. Automated PCB Manufacturability Checker
What to build: A tool that takes a KiCad board, runs design rule checks, and adds checks against the limits of a PCB manufacturer you pick (minimum trace width, clearance, drill size). Output a clear pass or fail report.
Tech stack: KiCad, kicad-cli pcb drc, Python, Streamlit for the interface.
What we checked: KiCad already has a design rule check, and the command line tool runs it with kicad-cli pcb drc and can write a json report. Do not rebuild DRC from scratch. Build on its output. If you want to read board data in Python, KiCad's developer docs say the old SWIG pcbnew bindings are deprecated since 9.0 and being replaced by the IPC API (use the kicad-python library). The dev docs currently plan to remove SWIG in KiCad 11, so check the release notes for the version you install. The IPC route needs KiCad running with its IPC server enabled.
Why recruiters like it: It shows you know PCB manufacturing constraints and can automate a tool engineers use daily.
Time to complete: 3-4 weeks.
Resume bullet: "Built a PCB manufacturability checker on KiCad's DRC engine with [N] extra fab-specific rules, tested on [N] boards, producing a pass/fail report that flagged [N] real issues"
Mechanical Engineering Projects (4 Ideas)#
14. Predictive Maintenance Dashboard for Rotating Machinery
What to build: A system that analyzes vibration data from motors or pumps, predicts bearing faults, and shows machine health on a dashboard.
Tech stack: Python, scikit-learn, Streamlit, SQLite. For data, start with the Case Western Reserve University Bearing Data Center, which has normal and faulty bearing data in Matlab format, sampled at 12,000 and 48,000 samples per second. Add your own accelerometer data after the model works.
What we checked: The MPU-6050 sensor from the February version is end-of-life. TDK's product page names the ICM-42670-P as the replacement. Cheap MPU-6050 boards are still sold and are fine for learning, but plan on a current sensor if you want to build on it. One modeling trap: if you cut one recording into random windows and split them across train and test, your accuracy will look amazing and mean little. Split by recording, load or bearing.
Why recruiters like it: It shows condition monitoring basics plus a real ML workflow, which is a useful mix for a mechanical student.
Time to complete: 3-4 weeks.
Resume bullet: "Developed a bearing fault classifier on the CWRU dataset using scikit-learn, reaching [X]% accuracy on held-out recordings, and built a Streamlit health dashboard with fault alerts"
15. CFD Simulation Comparison Tool
What to build: A tool that sets up and runs CFD simulations for simple geometries (pipe flow, airfoil, heat exchanger) with an open-source solver, then compares results across parameters.
Tech stack: OpenFOAM, Python for pre- and post-processing, ParaView for visualization, a simple web interface.
What we checked: There are two OpenFOAM lines, and tutorials for one do not always match the other. The OpenFOAM Foundation's latest is v14 (14 July 2026). OpenCFD's latest is v2606 (26 June 2026). Pick one, and only follow tutorials for that line.
Why recruiters like it: Most students only use ANSYS in labs. Running OpenFOAM yourself shows you can work with open-source tools and script a workflow. Only 8 India job titles in our data mention CFD (JobRise job data, 10 Nov 2026), and core mechanical hiring is poorly covered by our crawlers, so check the career pages of the companies you want.
Time to complete: 3-4 weeks.
Resume bullet: "Built a CFD workflow with OpenFOAM and Python automating mesh generation and solving for [N] geometries, with results checked against [published reference / analytical solution] within [X]%"
16. 3D-Printed Robotic Gripper with Force Feedback
What to build: A robotic gripper that picks up objects of different shapes and adjusts its grip using force sensors, with 3D-printed mechanical parts.
Tech stack: CAD in Autodesk Fusion or SolidWorks, a 3D printer from your college lab or a makerspace, Arduino, servo motors, force-sensitive resistors, Python for control logic.
What we checked: Autodesk gives students a free education licence for Fusion, renewed yearly and for educational use only. There is also a three-year free personal-use licence for non-commercial projects. Read which one you qualify for. We did not check SolidWorks student pricing, so ask your college.
Why recruiters like it: It joins CAD, manufacturing, sensing and control in one object you can hold up on a video call.
Time to complete: 4 weeks.
Resume bullet: "Designed and 3D-printed a robotic gripper with force-feedback control using Arduino and FSR sensors, grasping objects from [X] g to [X] g with adaptive grip pressure, tested on [N] object types"
17. Digital Twin of a Simple Mechanical System
What to build: A digital twin of a simple physical system (a pendulum, a gear train, or a spring-mass-damper). Build the physical system, add sensors, and run a live simulation that mirrors it.
Tech stack: Arduino with sensors (encoder, accelerometer), Python (SciPy and matplotlib) or MATLAB/Simulink if your college gives you a licence, serial communication for streaming, a simple dashboard.
A plain definition: A digital twin is a model that is fed live data from the real system and tracks it. If your simulation runs on its own and never sees sensor data, it is just a simulation. Make sure yours closes that loop and say so clearly in your README.
Why recruiters like it: The phrase "digital twin" gets thrown around a lot. A project that uses it correctly and shows the sensor-to-model loop stands out for that reason.
Time to complete: 3-4 weeks.
Resume bullet: "Created a digital twin of a spring-mass-damper system, synchronizing physical sensor data with a real-time Python simulation with [X]% measured deviation between physical and virtual behavior"
AI/ML Projects (4 Ideas)#
18. Indian Language Hate Speech Detector
What to build: A model that detects hate speech and offensive content in Indian languages (Hindi, Hinglish, Tamil, Bengali), with a simple web interface where users paste text and get a score.
Tech stack: Python, Hugging Face Transformers, an Indic-language model such as IndicBERT v2 (MIT licence, 23 Indic languages plus English), FastAPI, Streamlit or React.
What we checked, and what changed:
- No X scraping. The free X API tier is reported closed to new developers since February 2026, and paid reads are billed per post, per third-party pricing guides and another. Check X's own docs, then use published research datasets instead.
- Datasets: The FIRE HASOC track ran text tasks in English, Hindi and German across 2019-2024 and added Tamil and Malayalam in 2020. HASOC 2025 was about abusive memes in Hindi, Bengali, Gujarati and Bodo, and Tamil-English offensive text sits in a separate FIRE 2025 challenge. Check each dataset's licence and download terms. We did not verify a Bengali text dataset, so search for one before you promise Bengali.
- Hardware: Current Transformers needs Python 3.11+ and PyTorch 2.6+.
- Koo is gone. It shut down in July 2024.
Report results per language. A single overall score can hide a language where your model is weak.
Why recruiters like it: Content moderation is a real problem for any platform with Indian users, and handling code-mixed text is harder than English-only NLP. In our data, 129 India data roles are tagged NLP and 36 Hugging Face (JobRise job data, 10 Nov 2026).
Time to complete: 3-4 weeks.
Resume bullet: "Fine-tuned IndicBERT v2 on [dataset name] for Hindi/Hinglish hate speech detection, reaching [X] F1 on a held-out test set of [N] samples, served through a FastAPI endpoint"
19. AI-Powered Traffic Signal Optimization
What to build: A system that counts vehicles at an intersection from video, and adjusts signal timing based on traffic density. Test it in a simulator and compare against fixed-timer signals.
Tech stack: Python, an Ultralytics YOLO model for vehicle detection, OpenCV, the SUMO traffic simulator, Streamlit.
What we checked: Ultralytics lists YOLO26 as its latest model, YOLO11 as the mature option, and YOLOv8 as still supported for existing pipelines. Start with YOLO11 or YOLO26. Ultralytics code is AGPL-3.0, with a separate enterprise licence for commercial products, which is fine for a student project. SUMO is free (EPL-2.0 or GPL-2.0) and its current download is 1.28.0. For the comparison to mean anything, run both control methods on the same simulated traffic demand.
Why recruiters like it: It combines computer vision, optimization and simulation, and the before-and-after chart is easy to explain in an interview. In our data, 47 India data roles mention computer vision.
Time to complete: 3-4 weeks.
Resume bullet: "Developed a traffic signal optimizer using YOLO vehicle detection and SUMO simulation, reducing average vehicle wait time by [X]% versus fixed-timer signals across [N] simulated intersection setups"
20. Personalized Learning Path Recommender
What to build: A recommender that takes a user's current skills, career goal and weekly time, then suggests an ordered sequence of courses and projects.
Tech stack: Python, scikit-learn (TF-IDF and cosine similarity) or sentence embeddings, FastAPI, a React or Streamlit frontend. Build the course catalogue yourself as a CSV of 100-200 courses with skills, level and duration.
What changed: The February version scraped NPTEL, Coursera and YouTube and used collaborative filtering. Scraping those sites may break their terms (we did not check each one), and collaborative filtering needs a table of which users liked which items. You will not have that on day one. A content-based recommender works with zero users: match the skills a person wants to the skills each course teaches. Be upfront in your README about that, and add collaborative filtering later if real users show up.
Why recruiters like it: Recommendation is a common problem in product teams, and being honest about the cold-start problem shows more understanding than a fake accuracy number.
Time to complete: 3 weeks.
Resume bullet: "Built a content-based learning path recommender over a hand-curated catalogue of [N] courses using TF-IDF similarity and skill-gap ordering, evaluated by [N] test users rating [X]% of suggestions as relevant"
21. Document Summarizer for Indian Legal Texts
What to build: A tool that takes a long Indian court judgment and writes a short summary. Indian legal documents are long and dense, so a useful summary is worth something.
Tech stack: Python, Hugging Face Transformers (a seq2seq model such as BART or T5, or a long-context model), pypdf for PDFs, FastAPI, a simple web UI.
Datasets we confirmed: The IIT Kharagpur and IISER Kolkata team behind Law-AI/summarization published IN-Abs, with 7,130 Indian Supreme Court judgments and headnote-style summaries (7,030 for training, 100 for testing), plus IN-Ext with 50 expert-written extractive test pairs. They are listed on Zenodo and described in their paper. The same search turned up MILDSum, with 3,122 judgments and English plus Hindi summaries (EMNLP 2023). Read each licence before you reuse the data.
Watch for: Judgments are far longer than most models can read at once. Check your model's maximum input length, and split documents into chunks. Also add a clear note that the tool is not legal advice.
Why recruiters like it: Summarization on domain text (legal, medical, financial) is harder than on news, and working with a published benchmark lets you compare against known results.
Time to complete: 2-3 weeks.
Resume bullet: "Built an Indian legal judgment summarizer on the IN-Abs dataset using [model], with chunked long-document handling, scoring ROUGE-L [X] on the 100-document test split"
Data Science Projects (4 Ideas)#
22. Indian Startup Funding Analysis
What to build: An analysis dashboard of Indian startup funding by sector, city, year and investor, with a simple model for a question you can honestly answer from the data.
Tech stack: Python, Pandas, Plotly or Seaborn, scikit-learn, Streamlit.
What we checked: Crunchbase's Basic API is closed to new users, per its own docs. We could not confirm any free Tracxn reports, so that link is gone. The most used public file is the Kaggle Indian Startup Funding dataset, which a mirror describes as January 2015 to August 2017 with about 3,044 rows. That is small and old. Newer Kaggle uploads exist for 2021 and early 2022, so check their provenance and licence.
Honest take: With a dataset this size, a "will this startup raise a Series B" model will be weak. The cleaner project is the analysis and dashboard: which sectors and cities got funded, how rounds changed over time. If you do build a model, report a baseline next to it.
Why recruiters like it: It shows the full data workflow: collecting, cleaning, exploring, visualizing and explaining. In our data, Python appears on 2,674 India roles and SQL on 1,937.
Time to complete: 2-3 weeks.
Resume bullet: "Analyzed [N] Indian startup funding rounds ([years]) and built an interactive Streamlit dashboard with [N] visualizations showing sector and city funding trends, plus a baseline-compared model with [X] accuracy"
23. Real-Time Air Quality Monitoring and Forecasting
What to build: A dashboard of air quality across Indian cities, with a forecast for the next 24 hours.
Tech stack: Python, Pandas, Prophet or a simple baseline for time series, Plotly, Streamlit.
What we checked:
- Live data: The data.gov.in resource Real time Air Quality Index from various locations is described as hourly, with CPCB as the source. A tutorial says you need an API key from your data.gov.in account and that this API gives current values, not history. We could not open the official page directly, so verify the resource ID and fields yourself.
- History: OpenAQ's API is on v3 and needs an API key sent in the
X-API-Keyheader, and the site also points to an archive on AWS S3. We did not check how complete its India coverage is, so look at your cities first. - Data quality: CEEW notes that the live feed is not put through any censoring process. Expect spikes and gaps, and show how you cleaned them.
- Prophet: Its repo says it has been in maintenance mode since v1.4.0, with bug fixes only and no new features planned (the latest release in the changelog is 1.5.0, 5 Oct 2026). It still works fine for a baseline. Always compare against a naive "same as the same hour yesterday" forecast, because that is hard to beat for air quality.
Why recruiters like it: It covers API integration, messy real-world data, time series and a map, and the topic gives you something real to talk about.
Time to complete: 2-3 weeks.
Resume bullet: "Built an AQI monitoring and forecasting dashboard for [N] Indian cities from CPCB data and OpenAQ history, with a Prophet forecast beating a same-hour-yesterday baseline by [X]% MAPE on [N] days of holdout data"
24. E-Commerce Price Tracker and Fake Discount Detector
What to build: A system that tracks product prices over time and flags fake discounts, where the "original price" is inflated just before a sale.
Tech stack: Python, SQLite or PostgreSQL for price history, Streamlit, Telegram Bot API for alerts, simple statistics for anomaly detection.
What changed: The February version scraped Flipkart and Amazon India with Playwright or Selenium. Two problems:
- Flipkart's marketplace terms are reported to prohibit scraping and robots. That is a secondary source, so read the current terms yourself. We could not confirm Amazon.in's position.
- Amazon's old PA-API 5.0 has been deprecated and replaced by the Creators API. New credentials are needed, and access depends on meeting sales requirements for Associates accounts.
Safer routes: use the official Creators API if you can qualify, or log prices yourself for 20-30 products you actually follow by hand over a few weeks. A script that fetches retailer pages is still scraping, so check the site's terms and robots.txt first. The fake-discount logic is the interesting part, and it needs weeks of history anyway. Do not publish scraped data.
Why recruiters like it: It combines data storage, time series checks and alerting, and "is this discount real?" is a question everyone understands. It also shows you read terms of service before building.
Time to complete: 2-3 weeks (plus the weeks it takes to collect history).
Resume bullet: "Developed a price tracker logging [N] products over [N] weeks, using a rolling-median rule to flag [N] suspected inflated-MRP discounts, with Telegram alerts to [N] users"
25. IPL Match Outcome Predictor with Live Dashboard
What to build: A model that predicts IPL match outcomes from team, venue, toss and recent form, and a dashboard that updates the prediction ball by ball.
Tech stack: Python, scikit-learn or XGBoost, Streamlit, and ball-by-ball data from Cricsheet.
What we checked: Cricsheet offers free ball-by-ball IPL downloads as JSON (and YAML), and lists 1,243 IPL matches when we looked on 10 Nov 2026. Most of its data is under the Open Data Commons Attribution licence, so credit Cricsheet publicly. We found no official free live scores API, and the February post's "Cricbuzz/ESPN API" link was not one. Build the live dashboard by replaying past matches ball by ball from the Cricsheet files. Say "replayed" in your README, since that is what it is.
Do this right: Only use information available before each ball, or you leak the answer. Compare your model to simple baselines (for example, always pick the toss winner), because a model that "beats 50%" is not impressive on its own.
Why recruiters like it: Every interviewer in India follows cricket, and the project covers feature engineering, evaluation and a live-style dashboard.
Time to complete: 2-3 weeks.
Resume bullet: "Built an IPL match predictor on [N] Cricsheet matches using XGBoost with time-aware validation, reaching [X]% pre-match accuracy against a [X]% toss-winner baseline, with a ball-by-ball replay dashboard in Streamlit"
How to Present These Projects on Your Resume#
Building the project is half the work. Presenting it correctly on your resume is the other half. Here are the rules:
The Resume Bullet Formula
"[Action verb] a [what you built] using [tech stack] that [quantified outcome]."
Every bullet should have all four components. Let us break down an example with its blanks filled in:
"Developed a real-time collaborative code editor supporting 5 simultaneous users with 80 ms median sync latency, using React, Socket.IO, Yjs and Monaco Editor, deployed on Vercel."
(The numbers in that line are only an example of the format. Use your own measured ones.)
- Action verb: Developed
- What: real-time collaborative code editor
- Tech stack: React, Socket.IO, Yjs, Monaco Editor
- Outcome: simultaneous users, latency, deployed
Numbers Make Everything Better, If They Are Real
Recruiters skim resumes fast. The Ladders 2018 eye-tracking study put the average first look at 7.4 seconds. HR Dive's write-up does not say how many recruiters took part, so treat it as a rough guide, not a law. Numbers still catch the eye. Include at least one metric you measured yourself:
- Users who tested it
- Performance (latency, accuracy, processing speed)
- Scale (rows of data, requests, concurrent connections)
- Improvement over a baseline ("reduced X by Y%")
If your project has no real users yet, say so honestly: "tested by 5 classmates" is better than an invented figure, and an interviewer can ask you how you measured it.
Link Your Projects
For every project on your resume, include two links:
- GitHub repository (with a proper README explaining the project, how to run it, and screenshots)
- Live deployment URL (if it is a web app)
If you are not sure how to write a good GitHub README, we have a guide for that. A well-documented GitHub profile is becoming as important as the resume itself for tech roles. For wording, see how to describe portfolio projects on a resume.
ATS Optimization for Project Descriptions
Your project descriptions need to be ATS-friendly. This means:
- Include the exact technology names from the job description (React, not "a JavaScript framework")
- Use standard section headers ("Projects" or "Technical Projects")
- Avoid tables, columns, or graphics in the projects section
- Spell out acronyms at least once (Machine Learning (ML), Natural Language Processing (NLP))
Before submitting your resume, run it through JobRise's ATS checker against the specific job description. It will tell you which keywords you are missing and how to add them naturally.
Still Not Sure Which Project to Pick?#
If you are staring at these 25 ideas and feeling overwhelmed, here is a simple decision framework:
-
Pick the branch-specific project closest to the role you want. Applying for backend roles? Build the microservices URL shortener (#8) or the job alert bot (#6). Want a data science role? Build the air quality forecaster (#23) or the IPL predictor (#25).
-
Pick something you are genuinely curious about. You will spend 2-4 weeks on this. If you are bored by the topic, you will cut corners, and it will show.
-
Pick something you can demo in an interview. Projects with visual outputs (dashboards, visualizations, physical devices) are easier to talk about and more memorable to interviewers.
Still stuck? JobRise's ProjectBrainstorm tool can help. Tell it your branch, the role you are targeting, your current skills, and how much time you have. It will generate personalized project ideas with tech stack recommendations for your situation. You can also browse what employers are hiring for in our live jobs list.
Your project section is the one part of your resume you have complete control over. Grades are fixed. College name is fixed. But projects? You can start one today, finish it in 3 weeks, and change how recruiters see your application.
Stop building Library Management Systems. Start building something that makes a recruiter pause and think, "Wait, a fresher built this?"
Project ideas generator
Get project ideas that fit your branch, skills, and interests, and build things recruiters ask about.
Advertisement
Send this to whoever has the interview this week.