Research Topics in Computer Science and Engineering

How to read this list

Every entry is a starting point rather than a finished title. Before adopting one, narrow it to a specific setting, dataset or workload, then search IEEE Xplore and the ACM Digital Library for work from the last three years. If someone has already answered your exact question, change the condition, the data or the constraint. AI topics have a separate list, so this page concentrates on systems, security, data, networks and software engineering.

Cloud and distributed systems

1. Reducing cold starts in serverless functions — OpenFaaS, Kubernetes, Python
   Functions that sit idle respond slowly on the next call. Compare warm pools, snapshot restore and call-pattern prediction, measuring latency against the extra idle cost.
2. Carbon-aware job scheduling — Kubernetes, Python, public grid data
   Batch jobs can often wait for cleaner electricity. Build a scheduler that shifts deferrable work using grid-intensity data and measure the emissions saved against the delay introduced.
3. Autoscaling that resists traffic spikes — Kubernetes, Prometheus, Locust
   Threshold-based scaling reacts after users are already affected. Test short-term forecasting against the default autoscaler on replayed traffic traces, comparing missed response targets and resource waste.
4. Consistency trade-offs in multi-region databases — CockroachDB, Cassandra, YCSB
   Stronger guarantees add delay across regions. Benchmark an application under several consistency settings with simulated network latency and identify which operations can safely use weaker guarantees.
5. WebAssembly as a lightweight container alternative — Wasmtime, Docker, Rust
   Wasm runtimes promise fast startup and strong isolation. Compare them with containers on startup time, memory and throughput for realistic workloads, and report where each one wins.

Cyber security and privacy

1. Phishing detection for regional-language messages — Python, scikit-learn, IndicNLP tools
   Many filters are built on English text. Collect or synthesise Hindi and Marathi scam messages, then measure how well lightweight classifiers transfer and where they fail.
2. Detecting malicious packages in open-source registries — Python, static analysis, npm and PyPI metadata
   Typosquatted and compromised packages reach developers through normal installs. Analyse install scripts, metadata and name similarity to flag risky packages, and measure false alarms on popular ones.
3. Migrating web applications to post-quantum TLS — OpenSSL, liboqs, Nginx
   Newer key exchange methods use larger keys and handshakes. Measure handshake time and failure rates on slow mobile networks, then recommend practical deployment settings.
4. Intrusion detection on low-power IoT gateways — Raspberry Pi, Python, Zeek
   Heavy detection engines do not fit on small devices. Design a reduced feature set, test it on a public IoT traffic dataset and report accuracy alongside CPU and memory use.
5. Privacy leakage in mobile app permissions — Android, Frida, Python
   Apps often request more access than their features need. Build a tool that compares declared permissions with observed behaviour for a sample of apps in one category.

Data engineering and databases

1. Learned indexes versus B-trees on changing data — C++, Python, SOSD-style benchmarks
   Learned indexes look strong on read-heavy static data, but updates can hurt them. Benchmark both under mixed workloads and find the write ratio at which the advantage disappears.
2. Query tuning for vector search filters — PostgreSQL, pgvector, Qdrant
   Combining similarity search with ordinary filters can reduce recall or speed. Compare pre-filtering and post-filtering strategies across datasets with different filter selectivity.
3. Automatic data quality checks for pipelines — Apache Airflow, Great Expectations, Python
   Silent schema and distribution changes corrupt downstream reports. Build a monitor that learns normal column behaviour, inject known faults and measure how many it catches and how quickly.
4. Handling late events in stream processing — Apache Kafka, Apache Flink
   Events from unreliable networks arrive out of order. Compare windowing and watermark strategies on a delayed sensor stream, trading result accuracy against reporting delay.

Networks, edge computing and IoT

1. Task offloading between edge devices and cloud — iFogSim, Python
   Deciding where to run a task affects latency and battery life. Simulate a policy that adapts to link quality and compare it with always-edge and always-cloud baselines.
2. LoRaWAN coverage for rural monitoring — LoRa modules, ChirpStack, NS-3
   Long-range networks can serve farms and water systems, but terrain and interference matter. Model or field-test packet delivery and tune spreading factors to balance range and energy.
3. Comparing QUIC and TCP on unstable mobile links — Mininet, tc netem, Nginx
   Handovers and packet loss are common on mobile connections. Emulate these conditions and compare page load and video stall behaviour, identifying where QUIC offers a real benefit.

Software engineering and human-computer interaction

1. Reducing flaky tests in continuous integration — GitHub Actions data, Python, pytest
   Tests that pass and fail without code changes waste developer time. Mine the history of open-source projects, classify common causes and evaluate a predictor for likely flaky tests.
2. Quality and security of AI-assisted code — Semgrep, SonarQube, Python
   Generated code can compile yet contain weaknesses. Give a set of programming tasks to coding assistants, scan the output and compare it with human-written solutions on defined criteria.
3. Refactoring monoliths into microservices — Java, Spring Boot, graph clustering
   Choosing service boundaries is usually manual and subjective. Use call graphs and change history to propose boundaries for an open-source monolith, then evaluate coupling before and after.
4. Accessibility of Indian government and education portals — axe-core, Lighthouse, NVDA
   Screen-reader users still meet barriers on public services. Audit a defined sample of sites with automated tools and manual testing, then rank the most damaging recurring faults.

Turning a topic into a research question

A broad theme such as "cloud security" cannot be evaluated, while a question can. A good formula is: under a stated condition, does method A improve metric M compared with baseline B? For example, under simulated 4G packet loss, does QUIC reduce video stalls compared with TCP? The question immediately tells you what to build, what to measure and what counts as a result.

Check three things before committing. Can you get the data or a credible simulator? Can you run the experiment on hardware you can actually access? Is there a clear baseline to beat or compare with? A negative or mixed result is still valid research if the method is sound, so aim for a fair experiment rather than a guaranteed improvement.

• Undergraduate work: favour a working system with a clear comparison against an existing approach.
• ME or MTech work: expect a more defensible gap, a proposed method and a stronger evaluation.
• PhD-level interest: treat these as entry points for a wider literature review, since each one is too narrow on its own.

Getting guidance while you do the work

Road2Tech, based in Pune, mentors engineering students through this process, from reference papers and synopsis to algorithm details, testing reports and IEEE paper support. You remain the researcher and should be able to explain every decision in your viva.