BSc CSIT Final Year Project Topics (Tribhuvan University)
The BSc CSIT final year project fails on scope more often than on skill. A project proposed as a full platform and delivered as three broken screens marks worse than a narrow tool that works end to end and is properly evaluated.
Every topic below is sized for a small team over one academic year, and each names what data it needs — which is where CSIT projects most often stall, because a machine learning idea with no dataset is not a project.
Feasibility ratings here describe implementation risk, not marks. A Demanding project done well marks highest; a Demanding project abandoned in April marks lowest. Choose honestly against your team's actual ability.
What we help with
Scope review — An honest read on whether your proposed project is one year of work or three, before you commit to it in the proposal defence.
Proposal and documentation — Problem statement, objectives, methodology, system diagrams and Gantt chart in the format the CSIT project guideline expects.
System design — ER diagrams, DFDs, use case and sequence diagrams that actually match the system you are building.
Report writing — Chapter structure, testing documentation and result presentation for the final report and defence.
Answer the data question before the proposal defence
For every topic above, the first question your supervisor should ask is where the data comes from. For web and systems projects the answer is usually synthetic or self-generated, which is fine. For machine learning projects it is the whole project.
If your dataset does not exist yet, building and labelling it is legitimate work — but it must appear in your timeline as a phase with weeks attached, not as an assumption. Teams that plan four weeks of modelling and zero weeks of data collection are the teams that present an unfinished project in the final semester.
Check licensing too. A dataset that forbids redistribution may still be usable for your project, but you cannot ship it in your submission, and you should say so in the report rather than be asked about it at the defence.
What examiners actually look for
A working demonstration. Not a video, not screenshots — a system that runs in front of them. Everything else is secondary to this.
An evaluation with a baseline. A model reported at 94% accuracy means nothing without a comparison point. A system with no usability or performance testing has not been evaluated, only built.
Documentation that matches the code. ER diagrams and DFDs drawn at proposal stage and never updated are immediately visible, and they undermine trust in the rest of the report.
Honest limitations. A final chapter that states what does not work and why marks better than one claiming the system is complete, because examiners will find the gaps anyway.
Web applications
Web projects are the safest choice for a mixed-ability team — the stack is familiar, deployment is cheap, and a working system is demonstrable at the defence. Differentiate on the problem, not the technology.
Local produce marketplace connecting Nepali farmers to buyers
A genuine market failure with a clear user story on both sides, and a scope that stays manageable if you limit it to listing and matching rather than payments and logistics.
Data
Seeded catalogue data; optionally price references from Kalimati market published rates.
Method
Standard three-tier web application with role-based access; evaluate with task-completion testing on real users.
Feasibility
Straightforward
Complaint and grievance tracking system for a Nepali municipality
Municipal complaint handling is genuinely paper-based in much of Nepal, so the problem is real and a demo lands well with examiners.
Data
Synthetic complaint data; interview a ward office for realistic categories and workflow.
Method
Workflow engine with status transitions and SLA tracking; usability evaluation.
Feasibility
Straightforward
Blood donor matching and availability platform
Well-defined matching logic, a clear urgency dimension, and a social case that carries a defence well.
Data
Synthetic donor records; blood group compatibility rules are fixed and documented.
Method
Geospatial proximity search plus compatibility matching; notification subsystem.
Feasibility
Moderate
Academic plagiarism pre-check tool for college submissions
Constrained enough to be finishable and directly relevant to your own institution, which helps at defence.
Data
A local corpus of submitted documents, with permission and anonymisation.
Method
Shingling and Jaccard or cosine similarity; evaluate precision and recall against manually labelled pairs.
Feasibility
Moderate
Lost and found portal for a university campus
Small, honest scope with a real user base you can actually test on — which is worth more than an ambitious system nobody uses.
Data
Synthetic item data plus a pilot on your own campus.
Method
Text matching on item descriptions plus image upload; measure match success in the pilot.
Feasibility
Straightforward
Scholarship discovery and eligibility matching for Nepali students
The information genuinely is scattered, and rule-based eligibility matching is a defensible core contribution.
Data
Publicly listed scholarship criteria compiled into a structured dataset.
Method
Rule engine over a normalised criteria schema; evaluate against manually checked cases.
Feasibility
Moderate
Mobile applications
Mobile projects demo well but eat time on platform issues. Budget a month you did not plan for, and do not attempt iOS unless someone on the team already has a Mac.
Offline-first field data collection app for rural surveys
Connectivity is the actual constraint in rural Nepal, so offline sync is a real engineering contribution rather than a feature list item.
Data
Synthetic survey forms; model on a real NGO questionnaire if you can get one.
Method
Local store with conflict-resolving sync; evaluate under simulated intermittent connectivity.
Feasibility
Demanding
Public bus route and fare information app for Kathmandu Valley
Route information genuinely is tribal knowledge in the Valley, and the data collection itself is a defensible part of the work.
Data
Route and stop data you collect and digitise; document the collection method.
Method
Graph search over the route network for shortest and fewest-transfer paths.
Feasibility
Demanding
Medication reminder and adherence tracker
Small, well-bounded, and evaluable against a clear outcome measure.
Data
Synthetic schedules; a small pilot with consenting volunteers.
Method
Scheduling and notification subsystem; measure adherence rate in the pilot.
Feasibility
Straightforward
Nepali sign language learning app with gesture practice
Combines a real accessibility need with a technical core, and the scope stays sane if you limit the vocabulary set.
Data
A recorded gesture set for a limited vocabulary, produced with guidance from a signer.
Method
Video playback plus optional pose-based practice feedback; evaluate learning with a pre-post test.
Feasibility
Demanding
Campus attendance system using QR or geofencing
A common choice, so differentiate on the anti-spoofing analysis rather than on the basic feature.
Data
Synthetic class and roster data plus a live campus pilot.
Method
Time-bounded token generation with location validation; evaluate spoofing resistance explicitly.
Feasibility
Moderate
Earthquake preparedness and alert app for Nepali households
Nationally relevant, and the offline-content design decision is genuinely interesting to defend.
Data
Published preparedness guidance; seismic feed if a public API is available.
Method
Offline content store plus alert subscription; evaluate content retrieval under no-connectivity conditions.
Feasibility
Moderate
Machine learning and data science
The single question that sinks ML projects: where is the labelled dataset? If the answer is "we will create it", budget half your year for that and say so in the proposal. Nepali-language projects almost always need you to build the dataset.
Nepali news classification by category
Text is scrapeable, labels come from the source section, and the Nepali-language angle makes it a real contribution rather than a tutorial reproduction.
Data
Scraped Nepali news articles with section labels; document your scraping ethics and robots.txt compliance.
Method
TF-IDF with classical classifiers as baseline, then a transformer fine-tune; report precision, recall and F1 per class.
Feasibility
Moderate
Devanagari handwritten character recognition
A well-understood problem with an available public dataset, which makes it a safe choice that can still be pushed further.
Data
Public Devanagari handwritten character datasets.
Method
CNN baseline with augmentation; compare architectures and report a confusion matrix.
Feasibility
Straightforward
Nepali sentiment analysis on social media comments
Genuinely unsolved for Nepali and code-mixed Nepali-English, so honest results are publishable at student level.
Data
Collected public comments, manually labelled by your team with an inter-annotator agreement score.
Method
Baseline lexicon approach vs fine-tuned multilingual model; report agreement and F1.
Feasibility
Demanding
Crop disease detection from leaf images
Agricultural relevance and available public image datasets make this achievable, provided you are honest about domain shift to Nepali field conditions.
Data
Public plant disease image datasets; a small locally collected validation set is what lifts the project.
Method
Transfer learning on a pretrained CNN; evaluate separately on public and local images and discuss the gap.
Feasibility
Moderate
Air quality prediction for Kathmandu Valley
A real local problem with published monitoring data, and time series forecasting is well within CSIT coursework.
Data
Published air quality monitoring series for Valley stations, plus weather variables.
Method
ARIMA or LSTM forecasting; report error against a persistence baseline, which many student projects omit.
Feasibility
Moderate
Load shedding or electricity demand forecasting
Directly relevant to Nepal, and the modelling is tractable if you keep the horizon short.
Data
Published NEA demand and generation data where available; document any gaps honestly.
Method
Time series regression with seasonality terms; evaluate with rolling-origin validation.
Feasibility
Demanding
Systems, security and networking
These mark well with examiners who want depth over screens, but they demo poorly unless you plan the demonstration deliberately. Build a visualisation of what your system is doing.
Network intrusion detection using machine learning
Public labelled datasets exist, so unlike most ML topics the data problem is already solved for you.
Data
Public intrusion detection datasets such as those widely used in the literature.
Method
Feature selection plus classifier comparison; report false positive rate, not just accuracy.
Feasibility
Moderate
File integrity monitoring tool for Linux servers
A genuinely useful systems utility with a clear correctness criterion.
Data
No external data needed; test against controlled filesystem modifications.
Method
Hash-based baseline comparison with alerting; evaluate detection latency and false positives.
Feasibility
Straightforward
Secure document verification using digital signatures
Certificate forgery is a real problem in Nepal, which gives the project a defensible motivation.
Data
Synthetic documents; standard cryptographic libraries.
Method
Signature generation and verification workflow with a QR-based verification path; analyse the threat model explicitly.
Feasibility
Moderate
Campus network traffic analysis and reporting tool
Uses infrastructure you already have access to, and produces findings your own institution will care about.
Data
Captured campus traffic with explicit written permission and anonymisation.
Method
Flow aggregation and protocol breakdown with a reporting dashboard.
Feasibility
Demanding
Distributed file storage with redundancy
A core systems topic that lets a strong team show real depth rather than UI work.
Data
No external data; evaluate with synthetic workloads.
Method
Chunking with replication and recovery; measure recovery time and storage overhead under induced node failure.
Feasibility
Demanding
Vulnerability scanner for common web misconfigurations
Scoped to detection of known misconfiguration classes, this is achievable and genuinely instructive.
Data
Deliberately vulnerable test applications, run only in your own lab environment.
Method
Rule-based checks against a documented checklist; evaluate precision against known-vulnerable targets.
How big should a BSc CSIT final year project team be?
Follow your college's rule, which is usually two to four students. Whatever the size, divide by subsystem rather than by task type — a team where one person does all the coding and the others write documentation produces a weak defence, because only one person can answer questions.
Can I use an existing open-source project as a base?
Usually yes, provided you declare it clearly and your own contribution is substantial and identifiable. What is not acceptable is presenting someone else's system as your own build. State the base, state what you added, and the examiners will assess your contribution fairly.
Is a machine learning project riskier than a web project?
Yes, mostly because of data. A web project can be finished by working harder; an ML project with no usable dataset cannot be rescued by effort. If your team has not worked with data before, either pick a topic with a public dataset or choose a web or systems project.
What if the project does not fully work by the deadline?
Submit what works, document what does not, and be precise about why. A partially complete system with a clear-eyed limitations chapter marks considerably better than an overclaimed one that fails during the demonstration.
Do you build the project for us?
No. We help with scoping, proposal and documentation, system design diagrams, and reviewing your report. Writing the code is the part being assessed, and it stays with your team.
Is your service confidential?
100%. Your identity and academic work stay completely private and are never shared with your college or anyone else. We use secure, encrypted communication.
How do I get a quote?
Message us on WhatsApp at +977 9768768340 with your assignment brief and deadline, and we'll reply with a transparent quote — usually within a couple of hours.