Saif Ryan Gangaram

University of New Mexico, Department of Computer Science

Ph.D. in Computer Science, expected 2028

Focus: high performance computing, Big Data, AI/ML, signal processing, scientific data systems

Research advisor: Dr. Amanda Bienz

University at Buffalo, Institute for AI and Data Science

Master of Professional Studies in Data Sciences and Applications, 2023

University at Buffalo, College of Arts and Sciences

Bachelor of Arts in Mathematics, Computing and Applied Mathematics, 2021

Boundary-Correct Out-of-Core Deep Inference for Scientific Sensor Archives

Saif Ryan Gangaram, Adam Christopher Watts, and Dustin Yewell Harvey.

Accepted, IEEE International Conference on Big Data (IEEE BigData 2026).

An out-of-core execution contract for trained models over scientific sensor archives, using halo-aware slabs, hop-aligned short-time Fourier transforms, bounded PyTorch inference, and manifest validation. On the paper's 16-channel trace, slabbed processing matched the whole-file detection masks with a maximum absolute numerical difference of 4.2 × 10⁻⁶.

The official IEEE publication link will be added when available.

Research data scientist on the Test Engineering and Data Analysis team, working across AI and super-intelligence research, high-performance computing, big data, systems, signal processing, software engineering, statistics, and scientific data analysis.

• Conduct computer science research and write research papers, including the work accepted to IEEE BigData 2026.
• Develop scalable data acquisition, analysis, and compression workflows for experimental diagnostics and system validation.
• Apply time-series analysis, statistical modeling, numerical methods, AI/ML, and systems engineering to large, multi-axis sensor datasets to inform test design and performance evaluation.
• Build and integrate scientific and web pipelines using Python, C++, C, MATLAB, JavaScript, HTML, CSS, Echo, SQL, Bash, PowerShell, and Lean 4, with MPI and Slurm batch execution and JSON, HDF5, and NetCDF data products.
• Design signal processing, computer vision, deep learning, generative AI, and noise-mitigation workflows for complex experimental measurements.
• Support verification and validation through automated diagnostics, statistical quality checks, and data-driven readiness assessments.
• Optimize parallel workflows to reduce analysis latency, and refactor and integrate codebases with more than 250,000 lines of code.
• Develop and deploy deep learning training and inference workflows on HPC platforms for scientific datasets totaling more than 700 TB.
• Build full-stack agentic AI harnesses for analysis, software development, and management workflows, including multi-agent programming with human review.
• Advance secure data-analysis methods that support the reliability and safety of scientific and engineering systems.

• Developed scientific data pipelines, database-backed web tools, and automation workflows for aerospace and remote-sensing research settings.
• Integrated PostgreSQL, SQLite, HDF5, NetCDF, Python, C#, C++, Fortran, MATLAB, Flask, JavaScript, and dashboard tooling into analysis workflows.
• Built searchable data interfaces with visualization, selection, export, compression, schema inspection, and reproducible model-input generation features.
• Applied machine learning, anomaly detection, clustering, regression, simulation, image processing, and verification methods to large sensor, atmospheric, and telemetry-style datasets.
• Designed multithreaded communication and automation components using REST, TCP/UDP interfaces, synchronization primitives, and testable configuration workflows.
• Collaborated with multidisciplinary teams using Git, CI/CD practices, technical documentation, requirements analysis, integration testing, and iterative stakeholder feedback.

• Supported computer science labs and recitations for more than 45 students.
• Assisted with grading, exam administration, student mentoring, and occasional lecture support.

Protecting Tail Latency Under Contention

UNM CS 585 team project, Fall 2026. Comparing FIFO, FQ-CoDel, and CAKE queue management for small request-response messages competing with bulk TCP traffic. My responsibilities cover architecture, the paced UDP foreground workload, trial orchestration, and release integration. Implemented portable diagnostics, isolated Linux namespace setup and cleanup, workload accounting, and independent run-bundle validation. Primary collection and final policy comparisons remain pending.

github.com/srgangaram-swe/networks_project

Signal Foundry: Quantitative Research Workspace

Assembled AlphaForge and Signalattice in one repository while preserving source histories and independent dependency environments. Added a typed local API, bounded research workers, durable job records, and immutable diagnostic artifacts. The React/TypeScript Nexus workstation supports configuration, research comparisons, and evidence inspection. Paper operations require explicit authorization and qualified research; the system has no live-trading capability.

github.com/srgangaram-swe/SignalFoundry

Hyperion Z: Self-Hosted Agentic AI Operations Console

Built a self-hosted development console using Deno, TypeScript, and WebSocket. Its Autopilot turns a task into dependent tool-using sessions, streams their output, and presents the resulting Git diff. The browser includes a Monaco editor, tmux session control, saved SSH connections, file context, email drafting, and persistent agent configurations. A mock mode supports interface exploration without model API keys.

github.com/srgangaram-swe/Hyperion

Crucible: Local-First Training Data Refinery

Built a training-data platform with batch and streaming ingestion, quality checks and quarantine, exact and MinHash/LSH deduplication, dataset versioning, and a feature layer with point-in-time joins. Deterministic shards feed a reference transformer trainer with checkpoint resume. A metadata API, dashboard, and experiment harness support multi-seed studies of deduplication, data mixing, quality, and scale. The complete reference workflow runs offline on a laptop CPU; distributed backends are separately configurable.

github.com/srgangaram-swe/Crucible

Internal Agentic AI Workflow Harness

Built an internal agentic AI harness with Codex and Claude Code-style development assistance, multi-agent orchestration, approved internal LLM API integration, tool execution, enterprise collaboration context retrieval, meeting summarization, document generation, and email/workflow drafting. Designed the system for controlled internal environments, stakeholder-driven adoption, and phased expansion across a larger technical organization.

SynaptiCast: Causal Neural Telemetry Decoding and BCI Research Platform

Built a research platform for neural response decoding, causal forecasting, and simulated round-trip telemetry. The development benchmark separates subjects before training, fits calibration on validation data, and reports subject-level uncertainty, corruption stress tests, and latency. Classical baselines are evaluated beside neural models, including cases where the simpler decoder wins. A separate offline lab supports CNN and transformer decoding and self-supervised pretraining. These are software and simulation results, without claims of clinical or implanted-device performance.

github.com/srgangaram-swe/SynaptiCast

AxiomStack, DorkLLM, and DorkChat

Built a local language-model platform connecting a from-scratch decoder, evaluation, retrieval, and streaming chat. The development platform includes causal and grouped-query attention, rotary positions, KV caching, instruction tuning, and tests for model and sampling behavior. DorkChat exposes the model through a typed API and accessible browser interface, with document-grounded answers and reusable evaluation suites. Reference models use small public corpora and support local inspection and reproducibility.

github.com/srgangaram-swe/dork-llm

Signalattice: Calibrated Probabilistic Market Forecasting

Built a forecasting platform with chronological panel evaluation, calibrated ensembles, causal sequence models, and diagnostics for uncertainty, costs, delays, and liquidity constraints. Recent work includes a versioned DuckDB/Parquet feature store, bounded backfills, durable research jobs, and a read-only local evidence API. Research reports preserve source identity and explain each decision-readiness criterion. Synthetic and historical studies test the workflow without establishing live profitability.

github.com/srgangaram-swe/Signalattice

AlphaForge: Regime-Aware Quantitative ML Research Platform

Built quantitative ML research tooling with causal features and regime models, purged temporal validation, overfitting diagnostics, and out-of-sample costed backtests. Recent work adds constrained Markowitz optimization with independent KKT and objective checks, covariance and factor-risk snapshots, reconciled attribution, and reproducible hypothesis and trial records. The execution layer includes a self-financing simulator and a separately scoped C++17 order-book core. Rejected candidates and adverse stress results stay in the reports; no candidate is qualified for live trading.

github.com/srgangaram-swe/AlphaForge

Learning Systems Atlas: Evidence-Driven ML Systems Portfolio

Released a documented v1.0 research portfolio covering supervised and unsupervised estimators, a NumPy reverse-mode autodiff engine, PyTorch training, and reinforcement learning. Implementations include REINFORCE, DQN, and PPO, with separate validation and test streams. Reference experiments retain explicit seeds, versioned artifacts, model cards, and diagnostic plots. A three-seed comparison of nine model pairs with scikit-learn passed 25 of 27 held-out agreement checks and documented the two CART disagreements. Distributed ML and further algorithm/proof work remain planned.

github.com/srgangaram-swe/learning-systems-atlas

Conductor: Agentic AI Dashboard for Software Development

Developing a personal agentic AI dashboard tool for software development and workflow optimization. The project focuses on coordinating development tasks, surfacing project state, supporting debugging and documentation workflows, and helping developers move from intent to verified implementation.

PolytopeOS and the Polytope Compiler

Built a reproducible x86_64 UEFI reference path with a Rust loader and a separately linked freestanding kernel. The loader validates the kernel and transfers a bounded boot contract; the kernel checks that contract and emits allocation-free diagnostics. The workspace also includes a compiler lexer and a developer CLI. Boot evidence targets headless QEMU. Resource-aware scheduling, userspace, and physical hardware support remain future research work.

github.com/srgangaram-swe/polytope-os

Parallel Principal Component Analysis with MPI

Designed distributed PCA with rank-local sharding, global mean and covariance reductions, eigenpair broadcast, local projection, optional whitening, metadata logging, and scalability benchmarks.

Large-File Scientific ML Inference

Built an out-of-core inference workflow for scientific sensor archives using halo-aware slabs, bounded microbatches, core-trimmed NetCDF products, and manifest validation. The associated paper, listed above, has been accepted to IEEE BigData 2026. Its validation compares slabbed and whole-file outputs and checks coverage at slab boundaries.

Signal Processing and Workflow Reliability

Developed public-safe signal characterization workflows using spectral analysis, spectrograms, power estimates, stationarity tests, runtime tracing, and validation reports to improve confidence in noisy scientific data products.

Real-Time Human Emotion Recognition

Developed a CNN-based computer vision pipeline using public image datasets, augmentation, regularization, feature extraction, and real-time video inference.

Financial Risk Modeling

Developed probability-of-default models with gradient boosting, neural networks, feature engineering, Bayesian hyperparameter optimization, stress testing, and error analysis.

Scientific Database and Web Tooling

Designed database-backed research tools for search, plotting, export, schema inspection, data selection, and reproducible modeling workflows.

Quantitative Finance and Portfolio Optimization

Built a Python-based financial analysis system using portfolio objects, numerical analysis, and optimization methods to evaluate performance metrics and risk-aware allocation strategies.

Epidemiological SIR and Network Modeling

Implemented SIR and network-based epidemic models using public health datasets, numerical integration, and error analysis to study spread dynamics at regional scales.