Building agentic AI systems, retrieval pipelines, and security tools. Published IEEE research on multilayered honeypot architectures for network security. Led 30 interns to architect and deploy a production NLP chatbot. Built network traffic analysis tools processing millions of records for cyber threat detection. Interested in applied AI, retrieval and evaluation, cybersecurity, and full-stack development. Graduating May 2027 and currently looking for software engineering co-ops and new grad roles.
An autonomous agent that reads failing test output, localizes the fault, generates a candidate patch via LLM calls, writes the fix, and reruns the suite to verify the repair. Built with Python, Gemini API, and pytest, with bounded retries and validation gates so no unverified model output reaches the codebase. Evaluated on the QuixBugs defect benchmark, where near-perfect pass rates surfaced training-data contamination and moved evaluation to a harder held-out suite.
Hybrid retrieval over 640 vehicle patents combining BM25 keyword scoring, semantic embeddings, metadata filtering, and cross-encoder re-ranking. Benchmarked end to end, finding that retrieval pool size drives latency while metadata filters drive relevance.
A free, self-paced generative AI curriculum hosted on my site: five sessions from first principles to shipped products, five guided workbooks that each end in a working build, and an 88-section topic library. Static site with local progress tracking, no gating and no sign-up.
Full-stack country information platform built with React, TypeScript, and Go. Automated CI/CD pipeline with GitHub Actions deploying to AWS App Runner for zero-downtime releases, containerized with Docker, Redis caching, and Prometheus/Grafana monitoring.
Interactive cybersecurity analysis tool visualizing large-scale network traffic from the CSE-CIC-IDS2018 dataset (16M+ records). Profiled the pipeline with cProfile, found the bottleneck was file I/O rather than the model, and cut analysis time from 30 minutes to under 2 minutes using Python and Pandas while holding 95%+ detection accuracy.
Django application mapping US commercial sites on an interactive Leaflet map and estimating solar potential through the NREL PVWatts and Solar Resource APIs, built on a re-runnable pipeline with explicit failure states so bad data surfaces instead of disappearing.
Enhancing Network Security through a Multi-layered Honeypot Architecture with Integrated Network Monitoring Tools - First author, published at 2024 INDIACom, IEEE Xplore. Proposed a multilayered defense framework combining honeypots with network monitoring tools for cyber threat detection and mitigation.
Built CNN, RNN, and LSTM models using TensorFlow for text-based stress classification. Achieved 85% accuracy through ensemble majority voting on the ISEAR dataset.
Secure file storage application with end-to-end encryption (Fernet + SHA-256), blockchain-based integrity verification, and role-based access control built with Python, Flask, and SQLAlchemy.
| Category | Skills |
|---|---|
| AI & ML |       |
| Languages |        |
| Frameworks |      |
| Styling |     |
| Database |    |
| Cloud & DevOps |      |
| Services & Tools |     |
| Competitive Coding |    |
| Platforms |    |
Feel free to reach out if you want to collaborate on projects or discuss AI, cybersecurity, or software development.
- 🌐 Portfolio: tejashiv.github.io
- 📧 Email: tejas.shivaprasad29@gmail.com
- 💼 LinkedIn: linkedin.com/in/tejas-shivaprasad-a216a427a