AI & ML Infrastructure on Elixir/BEAM
Building core AI and ML infrastructure libraries primarily on the BEAM platform. Creator of Crucible Framework for LLM reliability research.
North Shore AI Ecosystem
A 7-tiered industrial ML infrastructure built on Elixir and the BEAM platform
Portfolio Ecosystem
Hexagonal architecture for Retrieval-Augmented Generation with 16 port specifications.
- portfolio_core - 16 Ports, ETS Registry
- portfolio_index - Pgvector, Neo4j, Broadway
- portfolio_manager - CLI, Multi-Provider
- portfolio_coder - Code Intelligence
Data & Agent Coordination
Complementary frameworks for data pipelines and multi-agent systems.
- flowstone - Asset-first data orchestration
- synapse - Multi-agent signal bus
- Lineage tracking & partition-aware execution
- Durable workflows with Postgres persistence
CNS Dialectical Engine
Thesis-Antithesis-Synthesis reasoning with critic-guided resolution.
- Proposer - Thesis generation, SNO extraction
- Antagonist - Contradiction detection, B1 gaps
- Synthesizer - Evidence-based resolution
- 5 Critics - Grounding, Logic, Novelty, Bias, Causal
Industrial Patterns
Organizing metaphors that encode architectural decisions into memorable concepts.
- Kitchen - Training infrastructure (crucible_kitchen)
- Cookbook - Training recipes (tinkex_cookbook)
- Forge/Anvil/Ingot - Data labeling stack
- Crucible - ML experimentation
Ingot Data Labeling
Sample factory library for generating, transforming, and computing measurements on samples
View on GitHub →Phoenix LiveView interface for sample generation and human labeling workflows
View on GitHub →Shared IR structs for the North Shore labeling stack (Forge/Anvil/Ingot) — typed datasets, samples, assignments, labels, artifacts, and evaluation runs for labeling workflows
View on GitHub →AI Agent Orchestration
Headless, declarative multi-agent orchestration framework with a domain-agnostic signal bus, workflow engine with Postgres persistence, and configurable agent runtime (ships code review domain).
View on GitHub →DSPEx - Declarative Self-improving Elixir | A BEAM-Native AI Program Optimization Framework
View on GitHub →Asset-first data orchestration for Elixir/BEAM. Dagster-inspired with OTP fault tolerance, LiveView dashboard, lineage tracking, checkpoint gates, and distributed execution via Oban.
View on GitHub →Multi-agent systems framework for the BEAM platform - build distributed autonomous agents with OTP supervision and fault tolerance
View on GitHub →FlowStone integration for altar_ai - AI-powered data pipeline assets with classify_each, enrich_each, embed_each helpers and unified telemetry
View on GitHub →Synapse integration for altar_ai - SDK-backed LLM providers for multi-agent workflows with automatic fallback, signal handlers, and workflow actions
View on GitHub →AI SDKs & API Clients
Elixir Interface / Adapter for Google Gemini LLM, for both AI Studio and Vertex AI
View on GitHub →An Elixir SDK for Claude Code - provides programmatic access to Claude Code CLI with streaming message processing
View on GitHub →Protocol-based AI adapter foundation for Elixir - unified abstractions for gemini_ex, claude_agent_sdk, codex_sdk with automatic fallback, capability detection, and telemetry
View on GitHub →
Ollixir provides a first-class Elixir client with feature parity to the official ollama-python library. Ollixir runs large language models locally or on your infrastructure via Ollama.
View on GitHub →Prompt and parsing utilities for Crucible and NSAI. Provides templating, schema validation, structured output parsing, and tool-call helpers for consistent LLM IO.
View on GitHub →Full-featured Elixir client for the Model Context Protocol (MCP) with multi-transport support, resources, prompts, tools, and telemetry.
View on GitHub →Shared LLM Actions for NSAI runtimes. Wraps PortfolioCore adapters with Jido.Action semantics and CrucibleIR.Backend input/output to centralize provider access.
View on GitHub →vLLM - High-throughput, memory-efficient LLM inference engine with PagedAttention, continuous batching, CUDA/HIP optimization, quantization (GPTQ/AWQ/INT4/INT8/FP8), tensor/pipeline parallelism, OpenAI-compatible API, multi-GPU/TPU/Neuron support, prefix caching, and multi-LoRA capabilities
View on GitHub →AI Infrastructure
A practical, multi-layered JSON repair library for Elixir that intelligently fixes malformed JSON strings commonly produced by LLMs, legacy systems, and data pipelines.
View on GitHub →High-performance, generalized process pooler and session manager for external language integrations. Orchestrates and supervises languages like Python and Javascript from Elixir.
View on GitHub →Compile-time Elixir code generator for Python library bindings. Declare dependencies in mix.exs, generate type-safe modules with introspected typespecs and docs. Deterministic git-friendly output, strict CI mode, streaming, and custom helpers. Runtime via Snakepit.
View on GitHub →Elixir RAG library with multi-LLM routing (Gemini, Claude, OpenAI, Ollama), GraphRAG, knowledge graphs, modular retrievers/rerankers, composable pipelines, pgvector integration, advanced chunking, and tool-using agents. Fork of bitcrowd/rag.
View on GitHub →Elixir SDK for the Tinker ML platform—LoRA training, sampling, and service orchestration built on OTP, Finch, and telemetry.
View on GitHub →Core Elixir library for AI agent orchestration - unified workbench for running, tracking, and orchestrating multi-provider LLM agents with sessions, workflows, RAG indexes, tool approvals, and cost tracking
View on GitHub →Elixir implementation of GEPA: LLM-driven evolutionary optimization using Pareto-efficient search for text-based systems. Features OpenAI/Gemini integration, BEAM concurrency, OTP supervision, 218 tests with 75% coverage.
View on GitHub →Unified API gateway for the NSAI ecosystem—authentication (JWT, API keys, OAuth2/OIDC), distributed rate limiting with burst allowance, circuit breakers, request tracing, Prometheus metrics, and service proxying. Production-ready with comprehensive observability and multi-tenant support.
View on GitHub →Service discovery and registry for the NSAI ecosystem—distributed registry with health checking, circuit breakers, multiple storage backends (ETS/PostgreSQL), PubSub event broadcasting, and comprehensive telemetry. Built on OTP with Horde-ready architecture for multi-node deployments.
View on GitHub →Interactive CLI and REPL for the NSAI ecosystem—unified interface to registry, gateway, jobs, experiments, datasets, embeddings, and metrics services; tab completion with bash/zsh scripts; configurable environments; JSON output mode for scripting; escript distribution. The cockpit for North Shore AI operations.
View on GitHub →Pure Elixir TikToken-style byte-level BPE tokenizer (Kimi K2 compatible).
View on GitHub →Chiral Narrative Synthesis workspace for Thinker/Tinker LoRA pipelines, semantic fact-checking, telemetry, and reviewer-ready CNS docs.
View on GitHub →Elixir port of tinker-cookbook: training and evaluation recipes for the Tinker ML platform.
View on GitHub →Elixir client for HuggingFace Hub—dataset/model metadata, file downloads, caching, and authentication. The BEAM-native foundation for HF ecosystem ports.
View on GitHub →Elixir port of HuggingFace's PEFT (Parameter-Efficient Fine-Tuning) library. Implements LoRA, AdaLoRA, IA3, prefix tuning, prompt tuning, and 30+ state-of-the-art PEFT methods for efficient neural network adaptation. Built for the BEAM ecosystem with native Nx/Axon integration.
View on GitHub →Hexagonal architecture core for Elixir RAG systems. Port specifications, manifest-based config, adapter registry, and DI framework. Enables swappable vector/graph/embedding/LLM backends via clean port/adapter abstractions. Hex.pm publishable.
View on GitHub →Production adapters and pipelines for PortfolioCore. Vector stores (pgvector, Qdrant), graph stores (Neo4j), embedders (OpenAI), Broadway pipelines, advanced RAG (Self-RAG, CRAG, GraphRAG, Agentic), multi-graph federation, and observability.
View on GitHub →AI-native personal project intelligence system - manage, track, and search across all your repositories with semantic understanding
View on GitHub →Schema & Validation
A powerful, flexible schema definition and validation library for Elixir, inspired by Python's Pydantic.
View on GitHub →Advanced typing and type validation mechanism for Elixir - runtime type checking and contract enforcement for BEAM applications
View on GitHub →Batch IR for standardized data interchange across training and inference backends. Covers text/token/chat/instruct batch types with encoding, validation, and portable schemas.
View on GitHub →Developer Tools
State-of-the-Art Introspection and Debugging System for Elixir/Phoenix Applications
View on GitHub →Revolutionary AST-based debugging and code intelligence platform for Elixir applications
View on GitHub →A Phoenix LiveView performance monitoring dashboard for tracking slow endpoints and database queries
View on GitHub →Code Intelligence Platform: Repository analysis, semantic code search, dependency graphs, and AI-powered code understanding built on the Portfolio RAG ecosystem. Features multi-language parsing (Elixir, Python, JavaScript, TypeScript), AST-aware chunking, call graph analysis, and intelligent code agents for review, refactoring, and documentation.
View on GitHub →Manifest-driven hexagonal core for generating Elixir SDKs and services with pluggable ports/adapters for transport, schema, retries, telemetry, streaming, and multipart.
View on GitHub →OTP & Distributed
Metaprogramming framework for automatic REST API generation from OTP operations
View on GitHub →Testing & QA
A battle-hardened testing toolkit for building robust and resilient Elixir & OTP applications.
View on GitHub →Observability
Lineage IR for cross-system traces, spans, artifacts, and provenance edges. Provides a shared event envelope and sink interface for consolidation across runtimes.
View on GitHub →Pachka-powered telemetry reporter for Elixir that batches client-side events, supports pluggable transports and :telemetry forwarding, and drains reliably on shutdown.
View on GitHub →Data & Databases
Vector embeddings service for Elixir—multi-provider support (OpenAI, Cohere, Voyage AI), intelligent caching with Cachex, batch processing with rate limiting, Nx-powered similarity computations, k-means/DBSCAN clustering, semantic deduplication, and ETS-based vector storage. Built for CNS and ML pipelines.
View on GitHub →Modern Elixir client for Weaviate vector database with health checks and friendly error messages
View on GitHub →Pure Elixir library for graph topology, TDA, and computational topology
View on GitHub →Composable regularization penalties for Elixir Nx. L1/L2/Elastic Net, KL divergence, entropy, consistency, gradient penalty, orthogonality. Pure Nx.Defn for JIT across EXLA/Torchx. Pipeline composition and Axon.Loop integration.
View on GitHub →Security
GUARDRAIL - MCP Security - Gateway for Unified Access, Resource Delegation, and Risk-Attenuating Information Limits
View on GitHub →Post-quantum cryptographic implementation of HQC (Hamming Quasi-Cyclic) - a NIST PQC candidate for quantum-resistant key encapsulation using code-based cryptography
View on GitHub →Research
Chiral Narrative Synthesis - Dialectical reasoning framework for automated knowledge discovery
View on GitHub →Utilities
Download high-quality audio from YouTube as MP3 files using Elixir. Features 104 music genres, duration filtering, per-genre caching, C++ DSP filters, and ML-based transient detection.
View on GitHub →Utility library and helper functions for Elixir development - common patterns, debugging aids, and productivity tools
View on GitHub →Elixir port of OpenAI's chz library - a powerful configuration management system for building composable, type-safe command-line interfaces with hierarchical configuration, environment variable support, and flexible argument parsing
View on GitHub →Client-agnostic multipart/form-data builder for Elixir with explicit file inputs, stream-first encoding, configurable form serialization, and adapters for common HTTP clients.
View on GitHub →Crucible Framework
A platform for conducting reproducible experiments on large language model reliability, built on Elixir/OTP.
Interactive Phoenix LiveView demonstrations of the Crucible Framework - showcasing ensemble voting, request hedging, statistical analysis, and more with mock LLMs
View on GitHub →Adversarial testing and robustness evaluation for the Crucible framework
View on GitHub →ML model deployment for the Crucible ecosystem. vLLM and Ollama integration, canary deployments, A/B testing, traffic routing, health checks, rollback strategies, and inference serving for Elixir-based ML workflows.
View on GitHub →ML feedback loop management for the Crucible ecosystem. Quality monitoring, data drift detection, model performance tracking, data curation, active learning, human-in-the-loop workflows, and continuous improvement for Elixir-based ML.
View on GitHub →CrucibleFramework: A scientific platform for LLM reliability research on the BEAM
View on GitHub →Experimental research framework for running AI benchmarks at scale
View on GitHub →Request hedging for tail latency reduction in distributed systems
View on GitHub →Industrial ML training orchestration - backend-agnostic workflow engine for supervised, reinforcement, and preference learning. Provides composable workflows, declarative stage DSL, comprehensive telemetry, and port/adapter patterns for any ML backend. The missing orchestration layer that makes ML cookbooks trivially thin.
View on GitHub →ML model registry for the Crucible ecosystem. Artifact storage, model versioning, lineage tracking, metadata management, model comparison, reproducibility, and integration with training pipelines for Elixir-based ML workflows.
View on GitHub →ML training orchestration for the Crucible ecosystem. Distributed training, hyperparameter optimization, checkpointing, model versioning, metrics collection, early stopping, LR scheduling, gradient accumulation, and mixed precision training with Nx/Scholar integration.
View on GitHub →Dataset management library for ML experiments—loaders for SciFact, FEVER, GSM8K, HumanEval, MMLU, TruthfulQA, HellaSwag; git-like versioning with lineage tracking; transformation pipelines; quality validation with schema checks and duplicate detection; GenStage streaming for large datasets. Built for reproducible AI research.
View on GitHub →Model evaluation harness for standardized benchmarking—comprehensive metrics (F1, BLEU, ROUGE, METEOR, BERTScore, pass@k), statistical analysis (confidence intervals, effect size, bootstrap CI, ANOVA), multi-model comparison, and report generation. Research-grade evaluation for LLM and ML experiments.
View on GitHub →HuggingFace Datasets for Elixir - A native Elixir port of the popular HuggingFace datasets library. Stream, load, and process ML datasets from the HuggingFace Hub with full BEAM/OTP integration. Supports Parquet streaming, dataset splitting, shuffling, and seamless integration with Nx tensors for machine learning workflows.
View on GitHub →Metrics aggregation and alerting for ML experiments—multi-backend export (Prometheus, InfluxDB, Datadog, OpenTelemetry), advanced aggregations (percentiles, histograms, moving averages), threshold-based alerting with anomaly detection (z-score, IQR), and time-series storage. Research-grade observability for the NSAI ecosystem.
View on GitHub →Training IR for reproducible ML jobs across Crucible and Kitchen. Defines model specs, adapters, learning config, checkpointing, validation, and resource envelopes to standardize training pipelines.
View on GitHub →Other Projects
Personal GitHub profile README with Elixir/AI projects and LLM reliability research
View on GitHub →Tech Stack
Specialties
- Distributed systems & fault tolerance
- AI/LLM infrastructure & reliability
- Functional programming & metaprogramming
- Statistical analysis & experimental design
- Developer tools & productivity