Gramian Consulting

• Migrate and refactor existing Python logic into Golang services owned by various other teams. • Rapidly learn and adapt to internal tools, systems, and processes to speed up migration. • Collaborate closely with engineers across teams to ensure smooth delivery. • Debug complex issues quickly and provide clear, actionable solutions. • Maintain high-quality, well-tested code with thorough validation. • Identify and implement improvements in our migration tooling.

• Solve advanced CUDA and GPU programming problems involving parallel computing and performance optimization • Review, evaluate, and improve AI-generated CUDA, C++, and Python code • Optimize GPU kernels for throughput, latency, memory efficiency, and resource utilization • Work with CUDA libraries and frameworks such as Thrust, cuBLAS, and cuDNN • Debug and resolve issues related to CUDA kernels, synchronization, and memory management • Develop high-quality technical prompts, solutions, explanations, and evaluations for AI model training • Collaborate with AI researchers, engineers, and evaluation teams • Stay up to date with the latest developments in CUDA, GPU architectures, and performance optimization techniques

• Develop, maintain, and extend high-performance, secure system software in C/C++. • Integrate and work with open-source projects such as OpenSSL, FFmpeg, glibc, systemd, tcpdump, libssh2, ImageMagick, and FreeRDP. • Design tools and infrastructure to support model training and evaluation workflows. • Build systems for suggesting code patches based on bug reports, issue descriptions, and vulnerability IDs. • Automate assessment of PRs and patch diffs for correctness and completeness. • Develop solution plans, set up reproducible environments, and evaluate test outcomes. • Collaborate with AI engineers on integrations involving Rust bindings and Python-based orchestration tools. • Analyze low-level system behavior using profiling, tracing, and debugging tools. • Write clear, maintainable code and documentation for internal systems and external libraries.

• Conduct **multi-turn conversations** with AI models and compare response quality across variants • Evaluate outputs for: - Accuracy and grounding - Personalization quality - Safety and responsible data usage • Test predefined **health & wellness scenarios** (fitness, activity tracking, nutrition, habits) • Identify issues such as: - Incorrect personalization - Hallucinations or reasoning gaps - Data misuse or privacy risks • Provide **clear, structured feedback and evaluation notes** • Maintain consistency and attention to detail across repeated evaluations

• Design vulnerable multi-component applications and security challenge environments across languages such as Go, Python, Node.js, or Rust • Develop realistic exploit chains combining multiple vulnerability categories and attack vectors • Build deterministic evaluation environments using Docker and automated validation tooling • Create security-focused test cases and verification logic for vulnerability detection and remediation workflows • Review and analyze AI-generated outputs to identify gaps in reasoning, security understanding, or exploit detection • Develop adversarial scenarios involving misleading documentation, obfuscated code, edge cases, and hidden attack paths • Model real-world vulnerability classes inspired by CVEs, bug bounty findings, and production security incidents • Ensure evaluation tasks remain scalable, reproducible, and resistant to contamination from public datasets • Collaborate with cross-functional teams working on AI evaluation, benchmarking, and automated testing systems

• Design and curate upper-undergraduate and graduate-level Chemistry problems with clear, fully worked solutions. • Create challenging, realistic tasks that expose failure modes in frontier AI systems. • Convert diagram-heavy or multi-part chemistry questions into self-contained, text-only problems with definitive answers. • Write rigorous explanations and verify scientific accuracy throughout the problem and solution. • Evaluate AI-generated responses, diagnose errors, and refine questions to meet difficulty targets. • Review Chemistry tasks created by others and improve clarity, correctness, and realism. • Collaborate with cross-functional teams to maintain consistent standards across large task batches.

• Design and run short multi-turn conversations (typically 1–5 turns) intended to test AI personalization behavior • Create prompts grounded in realistic personal scenarios to evaluate contextual understanding • Review AI responses to determine whether personalization is correctly applied • Check grounding quality to ensure the model does not invent unsupported claims about the user • Evaluate integration quality — confirming personal signals are used naturally (not forced or robotic) • Compare two responses side-by-side and determine which is more helpful, natural, and relevant • Write clear, structured rationales explaining rankings and referencing specific conversation turns • Verify debug information showing whether correct data sources were used • Maintain strict workflow hygiene (including deleting evaluation conversations when required)

• Design and solve challenging Physics problems to probe the limitations of large language models. • Create clear, high-quality, step-by-step solutions with well-articulated reasoning. • Collaborate with LLM researchers to align problems with evaluation goals, especially in areas where models struggle (e.g., abstraction, multi-step reasoning, symbolic manipulation). • Help define new evaluation benchmarks based on Physics curricula spanning early undergraduate to PhD-level topics.

• Design and develop power electronic circuits (e.g., DC-DC converters, LDOs, AC-DC supplies, automotive power systems) • Create and maintain schematics and PCB layouts using tools such as KiCad, Altium, or similar • Select and validate components based on electrical requirements and datasheet specifications • Evaluate circuit performance including efficiency, ripple, thermal behavior, and reliability • Apply best practices for PCB layout, including grounding strategies, routing, EMI mitigation, and DFM considerations • Prepare manufacturing deliverables such as BOMs, Gerber files, and pick-and-place data • Perform validation through simulations, calculations, and design rule checks (DRC) • Define structured validation methodologies covering electrical specs, component compliance, and design integrity • Establish evaluation criteria for PCB layouts, including DRC/LVS, thermal performance, and power integrity • Collaborate with engineering and cross-functional teams to improve design quality and evaluation processes

• Design and run short multi-turn conversations (typically 1–5 turns) intended to test AI personalization behavior • Create prompts grounded in realistic personal scenarios to evaluate contextual understanding • Review AI responses to determine whether personalization is correctly applied • Check grounding quality to ensure the model does not invent unsupported claims about the user • Evaluate integration quality — confirming personal signals are used naturally (not forced or robotic) • Compare two responses side-by-side and determine which is more helpful, natural, and relevant • Write clear, structured rationales explaining rankings and referencing specific conversation turns • Verify debug information showing whether correct data sources were used • Maintain strict workflow hygiene (including deleting evaluation conversations when required)