Currently a Founding Engineer at a stealth AI startup and Applied AI Researcher at Shankyaa, where I work on cryptographic agent security, prompt injection defences, and high-throughput ML pipelines. Deepening the security layer through an M.Sc. in Cybersecurity at BTU Cottbus-Senftenberg.
🏆 2nd place out of 6,000+ teams — RAISE Summit 2026, Paris | Factory Digital Twin with Physics-Informed Neural Networks & zero-trust sensor telemetry
🎯 Finalist — HackXplore Karlsruhe (ZEISS Track) | Cryptographic EEG/Motor BCI with Pedersen ZK-Proofs & 94% motor-intent accuracy at 250 Hz
🎯 Finalist — HackXplore Karlsruhe (LBBW Track) | Legatum — Intelligence-First Philanthropic Banking, 9-layer AI agent swarm turning philanthropic intent into on-chain verifiable impact
Open-source focus: AI agent security and the Model Context Protocol (MCP) ecosystem — finding and fixing attack surfaces in the infrastructure LLMs use to talk to the world.
| PR | What it fixes |
|---|---|
| googleapis/mcp-toolbox#3674 | Slowloris — adds ReadHeaderTimeout to prevent connection exhaustion on the Go MCP server |
| modelcontextprotocol/inspector#1732 | DNS-rebinding TOCTOU — pins resolved IPs so the proxy fetch can't be flipped to instance metadata |
| modelcontextprotocol/python-sdk#3141 | Unicode homoglyph spoofing — rejects tool names containing Cyrillic/Greek lookalikes and bidi chars |
| microsoft/PyRIT#2242 | Prompt injection — new MaliciousToolCallInjection attack strategy for agentic pipeline red-teaming |
| NVIDIA/garak#1981 | Homoglyph probes — tests LLM resistance to visual confusable attacks on tool names |
pip install mcp-scan
mcp-scan web_search wеb_search # catches Cyrillic е ≠ Latin eCLI scanner for MCP tool names — detects homoglyphs, bidirectional chars, and Unicode confusables → github.com/manjunathbhaskar/mcp-scan
