Matched Technical Projects & Opportunities
High-difficulty AI safety, concurrency audit, and domain-expert projects available for your skill profile.
Opportunities matched to your skills
View all 10 projects →Llama-3-70B Multi-Turn Jailbreak & System Prompt Evasion Red-Teaming
Construct complex multi-turn adversarial prompt trees attempting to breach safety alignment constraints in Llama-3-70B and Claude 3.5 Sonnet. Craft nested roleplay scenarios, base64 payload encodings, and cypher-shifted instructions to trigger policy violations, log vulnerability vectors, and calculate automated toxicity scores.
Rust Unsafe Lock-Free SPMW Ring Buffer & Loom Concurrency Audit
Evaluate and benchmark paired LLM outputs implementing a lock-free Single-Producer Multi-Worker (SPMW) queue in Rust utilizing std::sync::atomic ordering primitives (Acquire, Release, SeqCst). Audit unsafe pointer dereferences, identify memory model race conditions under Loom model checking, and write detailed performance comparison reports.
SEC Form 10-K DuPont Financial Ratio Extraction & RAG Hallucination Audit
Audit complex LLM RAG outputs extracting consolidated balance sheets, ASC 842 lease liabilities, and 3-step DuPont Analysis return on equity (ROE) ratios from 50 SEC Form 10-K filings. Verify mathematical derivations, detect subtle financial hallucinations in footnoted debt covenants, and construct ground-truth evaluation datasets.
CRISPR-Cas9 Off-Target CFD Scoring & Gene Regulation RLHF Preference Ranking
Rank and evaluate paired LLM candidate answers analyzing gRNA off-target cleavage probabilities using Cutting Frequency Determination (CFD) scores. Cross-reference variant frequencies from gnomAD v4 and Ensembl canonical transcript annotations to score biological accuracy and pathway impact.