Pranshu Dhyani
I build multi-agent AI systems
CSE student in New Delhi turning ideas into intelligent, polished software, from the model all the way to the interface.
A little about me
I’m a Computer Science and Engineering student based in New Delhi. I like projects where an idea has to survive contact with real data and real users.
This summer I trained at SSPL, a DRDO lab, working on retrieval-augmented multi-agent systems and a dynamic scenario prediction system. I build for hackathons and real problems across AI/ML and full-stack web.
“Ship something that works, then make it sharp.”
Tools I build with
Where I’ve worked
Practical Trainee, SSPL · DRDO
- Worked on retrieval-augmented multi-agent systems at the Solid State Physics Laboratory.
- Built a dynamic scenario prediction system as the training project.
- Documented the work in a detailed internship report.
Things I’ve built
Opinion Cascade Simulator
A multi-agent LLM system where 6–8 persona agents simulate opinion dynamics in a turn-based loop. Grounded with RAG and a two-tier memory (episodic SQLite + semantic Chroma), streamed live to a D3 visualization, and run locally with Ollama.
UEBA Insider-Threat Detection
Detects anomalous user behaviour on the CERT r4.2 dataset with Isolation Forest, streaming alerts live to a SOC-style dashboard.
OrderStream
An order-event pipeline: a FastAPI producer publishes to Kafka, a consumer persists to SQLite. Containerised with Docker Compose, tested with pytest, shipped through GitHub Actions to AWS EC2.