rishabh sinha

Google · Santa Clara

Software engineer · ML systems

I build AI
systems that move.

I work on low-latency inference, retrieval, and reliable agentic systems — currently at Google.

01 / Work

Experience

Production ML and distributed systems work across model serving, retrieval, agent orchestration, and scientific computing.

  1. Sunnyvale, CA

    Google

    Software Engineer

    • Delivered Gemini intent-encoder post-training at 100+ QPS and sub-25 ms p99 with QLoRA/SFT on four TPUs.
    • Lowered RAG retrieval p99 by 9% through fan-out DAGs and concurrent embedding-memory reads.
    • Improved Chat search grounding through planner verbalization and context-service agent orchestration.
    • Hardened serving reliability with safer Borg/Kubernetes rollouts, telemetry, and cell-level rollback.
  2. Seattle, WA

    Amazon Web Services

    Software Development Engineer

    • Architected a Bedrock AgentCore and LangGraph multi-agent orchestrator, cutting inference latency 40% with semantic caching.
    • Built a Kubernetes static-analysis engine around scheduler bin-packing, increasing validation throughput by more than 300%.
    • Improved agentic code-generation success by 25%+ with memory, sandboxed execution, and eval-driven repair loops.
    • Raised agent reliability by 30%+ through deterministic evals, trace replay, and guarded tool execution.
  3. Seattle, WA

    Amazon Web Services

    Software Development Engineer Intern

    • Built a fault-tolerant OAuth-backed integration for synchronizing resource data across 150+ AWS services.
    • Reduced on-call ticketing by 35% through resilient resource observability.
    • Implemented SigV4 authentication and 5 ms LDAP authorization using configuration caching and Guice.
  4. Gaithersburg, MD

    National Institute of Standards and Technology

    Software Engineering Intern

    • Accelerated signal processing by 85% with multiprocessing, NumPy SIMD operations, and parallel execution.
    • Reduced runtime by 45 seconds through thread pooling, profiling, and C++ hot-path optimization.
    • Identified 14 bottlenecks and configured PyCUDA kernels and AWS autoscaling for GPU workloads.
  5. College Park, MD

    University of Maryland Applied Research Laboratory

    Machine Learning Research Intern

    • Cut Lorenz Attractor model runtime by 20 seconds with NumPy, Dask, and Bash automation.
    • Improved prediction F1 by 15% using KNN, Truncated SVD, Ball Trees, and SHAP.
    • Reduced system capture size by 25% with adaptive time-delay embedding built on PyEMD.

02 / Selected builds

Projects

Measured systems work: throughput, tail latency, scaling, and reproducible evaluation.

  1. 01 SuperRed Composable AI red-teaming across attackers, targets, benchmarks, and explicit threat models. Python · AI Security
  2. 02 Low-Latency LLM Inference 120 tokens/s on Llama-3-8B with 40% lower p99 latency at 32-request concurrency. vLLM · C# · gRPC
  3. 03 RL Hyperparameter Tuner PPO-driven learning-rate tuning for ResNet-18, validated against a 910 img/s JAX ceiling. PPO · PySpark · JAX
  4. 04 SLURM Vision RAG LLaVA-1.5-7B fine-tuning on four A100s with a 1M-vector retrieval index. PyTorch · Slurm · RAG

03 / Research & OSS

Research

Current work on evaluating and understanding the behavior of AI systems.

  1. Preprint

    SuperRed: An AI Red-Teaming and System Security Framework

    Simon Sure · Sarthak Munshi · Rishabh Sinha · Zhun Wang · Dawn Song · ICLR 2027 submission

    A composable framework separating attackers, systems under test, security claims, and controllers. Security domains make adversary trust boundaries explicit, enforced, and sweepable across experiments.

    35
    modules at launch
    4
    swappable components
    Project Source
  2. Preprint

    Latent-State Auditing for Tool-Using Language Agents

    Rishabh Sinha · UC Berkeley

    Adds activation-derived evidence immediately before agent tool use. Learned NLA-text and raw-activation monitors flagged all 22 generated unsafe actions before execution in the primary AgentDojo-grounded evaluation.

    22 / 22
    unsafe actions flagged
    2.0
    mean steps of lead time
    Code & artifacts
  3. NeurIPS submission

    A Nonzero Final Learning Rate Floor Improves Transfer of Agent-Discovered Schedule Tweaks

    Rishabh Sinha · NeurIPS 2026 submission

    Tests whether an AutoResearch-discovered schedule change survives controlled decomposition, five-seed reruns, and nearby depth transfer. A 0.05 final learning-rate floor—not warmup—was the robust component in the fixed-budget single-GPU harness.

    4 / 5
    base seeds improved
    5 / 5
    depth-10 seeds improved
    0.000877
    base mean BPB gain
    0.001570
    depth-10 mean BPB gain
    Request preprint

04 / Education

Education

Computer science, mathematics, data systems, and computational finance.

2026

University of California, Berkeley

Master of Information and Data Science

Part-time professional program GPA 4.0

2025

University of Maryland, College Park

Bachelor’s in Computer Science and Mathematics

Minor in Computational Finance GPA 3.98

05 / Toolkit

Skills

A focused toolkit for production ML, distributed infrastructure, and performance work.

Languages
Python, C++, Java, C, C#, Go, Rust, TypeScript, JavaScript, SQL, Scala
ML & compute
PyTorch, JAX, vLLM, Ray, NumPy, PySpark, OpenCV, CUDA, TPUs
Systems
Kubernetes, Docker, Linux, Kafka, gRPC, AWS, Slurm, Prometheus, OpenTelemetry
Web & data
React, Node.js, FastAPI, Flask, PostgreSQL, Pinecone, Grafana