# Yudhister Singh > ML Systems Architect | Applied Scientist | Researcher | Co-Founder Location: San Francisco, California, United States Profile: https://flows.cv/yudhister Operating on scaling-law asymptotics and post-attentional inductive biases. Designing sparse, retrieval-conditioned, multimodal foundation systems under compute-optimal regimes and beyond canonical attention regimes - and adjacent frontier AI research. Frontier Stack: • Frontier LLMs: sparse MoE routing, hierarchical experts, speculative decoding, KV-cache compression, FlashAttention-2/3, sliding-window + recurrent hybrids • Post-Transformer sequence models: selective SSMs (Mamba-class), Hyena operators, RWKV, linear attention kernels, implicit state discretization • Alignment & optimization: RLHF, DPO, IPO, RLAIF, reward modeling, constitutional AI, scaling-law extrapolation, capability phase transitions • Long-horizon reasoning: tree-of-thought, graph-of-thought, tool-augmented agents, program synthesis loops, self-reflection policies • Agentic systems: autonomous planners, memory-augmented agents, vector-episodic recall, multi-agent coordination protocols • RAG at planetary scale: hybrid lexical–semantic retrieval, cross-encoder re-ranking, HNSW/IVF-PQ, embedding space whitening, retrieval calibration • Multimodality: vision-language alignment, diffusion transformers, video-text temporal grounding, speech-LLM coupling, cross-modal contrastive pretraining • Generative paradigms: diffusion models, rectified flows, consistency models, latent world models • Model efficiency: QLoRA, DoRA, low-rank factorization, distillation, structured pruning, weight sparsification, 4-bit/FP8 quantization, kernel fusion • Distributed training: ZeRO-3, FSDP, tensor/pipeline parallelism, gradient checkpointing, topology-aware scheduling • Inference systems: CUDA graph capture, Triton kernels, operator fusion, batching heuristics, microsecond-level latency budgets • Continual & federated learning: parameter interpolation, catastrophic forgetting mitigation, adaptive fine-tuning • Safety & robustness: adversarial resilience, red-teaming automation, interpretability probes, activation steering • Retrieval-conditioned cognition, embedding manifold geometry, anisotropy correction, representation collapse mitigation Research vector: scaling-law efficiency frontiers, architectural sparsity, adaptive state dynamics, emergent reasoning thresholds, alignment-constrained generalization, and post-attention intelligence substrates. Translating frontier literature into ML-native, distributed infrastructure. Mentoring research-grade engineers. Co-founding AI systems from first principles. ## Work Experience ### Senior Software Engineer - ML Systems @ Databricks Jan 2023 – Present ### Co-Founder @ Stealth AI Startup Jan 2026 – Present ### Machine Learning Researcher - Applied Scientist @ SLAC National Accelerator Laboratory Jan 2019 – Present | California • Predicted similarities for 3D tomograms that reduced false positives and improved accuracy to 64% in detecting outliers, using contextual learning, DoG (difference of Gaussian) and multivariate statistical analysis • Produced script that can automatically find repeating features in cryo-electron tomograms with the use of Machine Learning ### Senior Software Engineer - ML Systems @ Airbnb Jan 2021 – Jan 2023 ### Senior Software Engineer @ Verizon Jan 2021 – Jan 2023 ### Full Stack Developer @ Tech Mahindra Jan 2017 – Jan 2018 | Mumbai, Maharashtra, India • Designed and built systems to ingest and transform data using data mining, text analysis and query processing of over 100TB search and video logs using Amazon's Deep Scalable Sparse Tensor Network Engine (DSSTNE) Implemented IFCM in Python using triangular fuzzy sets with inputs in .csv and image formats, resulting in 78% accuracy • Created a user centric platform using Python that uploads Raspberry Pi energy information onto the internet server using Cloud storage. The platform uses Thingspeak that follows MQTT (Message Queuing Telemetry Transport) protocol • Developed cross-collaboration website for startups to support conference room booking, raising support tickets, and monitoring KPIs – using React and Redux as front-end, Node.js as backend and developed on Kubernetes ### IoT Engineer Intern @ Clove Technologies Jan 2017 – Jan 2017 | India Developed Watt Swot smart energy meter based on IoT algorithm to track, map and analyze the power consumption in both residential and commercial facilities ### Software Developer @ Aditya Birla Group Jan 2015 – Jan 2016 Launched the use of online and offline feedback, sentiments, and hashtags using Apache Spark MLlib and Singa within marketing campaigns to realize marketing targets. ### Software Developer @ Sony Jan 2014 – Jan 2014 | Mumbai Area, India Integrated the local office front end (ReactJs, jQuery, Ajax) and back end (Node.js, Ruby on Rails, Python) with that of head office in Australia’s using ReactJs, jQuery and Ajax to elevate customer’s UI/ UX experience ### Content Developer @ Shikshalaya Jan 2014 – Jan 2014 | Mumbai Area, India ## Education ### Master of Science - MS in Computer Science University at Buffalo ### Bachelor of Engineering - BE in Computer Science University of Mumbai ## Contact & Social - LinkedIn: https://linkedin.com/in/yudhistersingh - GitHub: https://github.com/Yudhister1 --- Source: https://flows.cv/yudhister JSON Resume: https://flows.cv/yudhister/resume.json Last updated: 2026-03-22