Team: Engineering — AI / Research

Start: ASAP

VulSight is a Web3 security firm helping teams ship safely across Move, Rust, and EVM ecosystems. We’ve delivered hundreds of security assessments in Web2 and Web3 and are now building AI-powered auditing tools that catch bugs earlier, accelerate reviews, and raise the bar for protocol safety.

We’re hiring an AI Tech Lead to own our AI roadmap end-to-end: research, prototyping, model/productization, and team leadership. You’ll build systems that reason over smart-contract codebases (Move, Solidity), diffs, on-chain data, and audit reports to surface vulnerabilities, generate tests, and assist human auditors. Start hands-on; quickly grow and lead a small team (3–6) on-site in Islamabad.

• Own the AI stack: model selection/finetuning, inference infra, evals, and guardrails for code-intelligence use cases.
• Ship auditor-grade copilots: code reading, invariant mining, property/test generation, patch suggestions, and risk summaries.
• Blend static + dynamic analysis with LLMs: integrate Static Analyzers, Foundry, fuzzers, and custom CFG/AST pipelines.
• Build retrieval over code & chain data: embeddings, chunking, and RAG for large repos, on-chain state, and historical diffs.
• Create robust evals: curate gold-standard benchmarks from public bugs, past audits, and synthetic cases; track precision/recall + business KPIs.
• Productionize: latency-aware inference (vLLM/TensorRT), cost controls, caching, and safe rollout processes.
• Lead & mentor: hire, coach, set quality bars, and establish a rapid, experiment-driven culture.

• 5–8+ years in software/ML; 2–3+ years building or leading ML/LLM products end-to-end.
• Strong coding skills in Python plus one of Rust/TypeScript/Go
• Deep experience with LLM systems (prompting/tool use, RAG, finetuning/LoRA, evaluation).
• Comfortable with PyTorch/JAX, distributed training/inference (Ray/vLLM), vector DBs, and data pipelines.

Nice to Have

• Security/compilers/program analysis (AST/CFG/SSA, SMT solvers, fuzzing).
• Web3 experience (Solidity/Foundry, Move on Aptos/Sui), reading on-chain data.
• Model distillation/quantization and GPU inference ops.

• Real impact: Protect user funds and prevent headline-level incidents.
• Ownership: Influence product, research directions, and the team you’ll lead.
• On-site excellence: High-focus environment in Islamabad and a builder culture.
• Growth:

• On-site in Islamabad (relocation support within Pakistan if needed).
• Generous research compute budget.