███╗ ███╗██╗ ██╗███╗ ██╗███████╗███████╗██████╗
████╗ ████║██║ ██║████╗ ██║██╔════╝██╔════╝██╔══██╗
██╔████╔██║██║ ██║██╔██╗ ██║█████╗ █████╗ ██████╔╝
██║╚██╔╝██║██║ ██║██║╚██╗██║██╔══╝ ██╔══╝ ██╔══██╗
██║ ╚═╝ ██║╚██████╔╝██║ ╚████║███████╗███████╗██████╔╝
╚═╝ ╚═╝ ╚═════╝ ╚═╝ ╚═══╝╚══════╝╚══════╝╚═════╝
Data & AI Engineer — AI security tooling, LLM evaluation harnesses, full-stack products
Islamabad, Pakistan · @turing · muneebs.space · ORCID 0009-0001-2042-7656
muneeb@horizon ~ $ whoami
Muneeb Anjum — Data & AI engineer, researcher, author.
muneeb@horizon ~ $ cat focus.txt
> LLM security prompt injection, quantization-induced attack surface
> Evaluation benchmarks that report what aggregate accuracy hides
> Quantization what survives compression, and what silently flips
> Product engineering Chrome extensions, MCP servers, self-hosted tools
muneeb@horizon ~ $ tree research --depth 1
research/
├── quantized-guardians accuracy-neutral decision churn in compressed detectors
├── smart-contract-verdict-churn verdict churn in quantized smart-contract auditors
├── sentinel-llm five-layer defense framework for adversarial prompts
└── tokenization-attack-kit 11 Unicode-level attack families + detector| Paper | What it measures | Status |
|---|---|---|
| Quantized Guardians — Accuracy-Neutral Decision Churn in Compressed Prompt-Injection Detectors | F1/AUROC hold steady while which inputs get blocked flips | DOI |
| Same Accuracy, Different Verdicts — Quantization-Induced Verdict Churn in Small LLM Smart-Contract Auditors | 175 Solidity contracts, F16 → Q8_0 → Q4_K_M, up to 45.7% verdict flips | DOI |
| SENTINEL-LLM — Practical Multi-Layer Defense for Adversarial Prompts | 7 PoC prompts, 32 real attacks (JailbreakBench / HarmBench / OWASP) | preprint + code |
| TokenAttackKit — Unicode-Level Tokenization Attack Detection | 11 attack families, 0.98 F1 heuristic detector, CPU-only | code |
muneeb@horizon ~ $ ./projects --sort recent --limit 6| Project | Stack | Description |
|---|---|---|
localbench |
Python FastAPI |
Benchmark console for local LLM servers — TTFT, prefill/decode tok/s across 1K–64K contexts, concurrency sweeps, GPU telemetry |
pin-saver |
JavaScript MV3 |
Download a whole Pinterest board as ZIP or PDF — zero dependencies, live on the Chrome Web Store |
linkedin-mcp-pro |
Python MCP Playwright |
MCP server exposing 54 LinkedIn tools to any agent runtime |
llm-quantization-benchmarks |
Python |
Accuracy delta, per-input decision churn and per-layer sensitivity across FP32/BF16/INT8/INT4 |
agentic-rag-pipeline |
Python |
Hybrid BM25 + dense retrieval, reciprocal rank fusion, query rewriting, grounding checks |
tinyllm-eval-suite |
Python |
Reproducible eval harness for small models — task suites, scoring, regression tracking |
muneeb@horizon ~ $ ls ~/engineering
speech-to-text-pipeline time-series-anomaly fraud-detection-system
neural-architecture-search mlops-experiment-tracking synthetic-data-generator
multimodal-vlm-training medical-imaging-diagnosis autonomous-rl-agent
edge-vision-tensorrtmuneeb@horizon ~ $ cat ~/books.txt
Practical AI Security Series — Muneeb Anjum
[01] Quantum Computing with AI Kindle · Paperback
[02] LLM Security and Prompt Injection Kindle · Paperback
[03] AI and Blockchain Kindle · Paperback
[04] AI and Cyber Security Kindle · Paperbackmuneeb@horizon ~ $ cat stack.toml
[core]
python = "primary — PyTorch, FastAPI, numpy"
js = "TypeScript, Chrome MV3, React"
infra = "Linux, systemd, Nginx, Docker, Cloudflare"
[ai]
models = "transformers, llama.cpp, vLLM, quantization tooling"
agents = "MCP, tool-calling, RAG, evaluation harnesses"
muneeb@horizon ~ $ uptime
writing code · measuring what it actually doeswriting code · measuring what it actually does
muneebs.space · GitHub · ORCID · LinkedIn