A curated monorepo of tested software projects across applied AI, reinforcement-learning post-training, machine learning, Rust protocol design, data engineering, and cloud security.
Every featured project includes its own setup instructions, validation path, and explicit scope boundaries. The root CI currently covers 17 projects: 10 Python projects, 6 Rust crates, and 1 Terraform configuration.
| Area | Recommended entry point | Why it is useful |
|---|---|---|
| LLM systems | RAG Assistant | Modular retrieval and generation, FAISS with a NumPy fallback, retrieval evaluation, and optional cross-encoder re-ranking |
| RL post-training | GRPO Minimal and Regularized Operator Zoo | A shared training loop for RFT, Online RFT, and GRPO+OS, backed by tested regularized operators |
| Rust systems | Cross-Chain Atomic Bridge | HTLC commit and reveal, rollback, conservation checks, native and wrapped asset accounting, and integration tests |
| Data engineering | Sales Data ETL | SSIS delivery artifacts plus a portable Python and SQLite reference path for CI verification |
| Browser demos | AI Playgrounds | Twelve bilingual interactive AI applets in a separate repository with no installation required |
The checked-in five-case smoke benchmark produced the following CPU result:
| Configuration | recall@3 | MRR |
|---|---|---|
| Embedding retrieval | 1.000 | 0.900 |
| Cross-encoder re-ranked | 1.000 | 1.000 |
| Observed delta | +0.000 | +0.100 |
The re-ranker moved one relevant result from rank 2 to rank 1. This is a reproducible smoke measurement, not evidence of a general quality lift; issue 18 tracks the larger discriminative benchmark.
On a synthetic verifiable-reward task, averaged across three seeds and 1,200 steps:
| Method | Final accuracy |
|---|---|
| RFT | 0.181 |
| Online RFT | 0.945 |
| GRPO+OS | 0.990 |
The project reproduces the qualitative ordering RFT < Online RFT < GRPO+OS associated with Figure 5 of Shao et al. 2024. It does not claim the paper's absolute benchmark values.
A verified run of the operator library produced:
- maximum policy-to-gradient residual of
1.8e-07; - maximum conjugate-identity difference of
2.2e-16.
See RESULTS.md for the recorded run context and regeneration commands.
| Project | Focus |
|---|---|
| RAG Assistant | Chunking, embeddings, vector search, cited generation, retrieval metrics, Flask serving, Docker, and optional cross-encoder re-ranking |
| Agent Toolkit | ReAct-style agent loop, typed tool registry, deterministic test LLM, append-only traces, and guarded built-in tools |
| LLM Eval Harness | JSON and YAML evaluation suites, exact and regex grading, embedding similarity, LLM-as-judge, and standalone HTML reports |
| NLP Text Summarization CLI | Async batching, concurrency limits, API-key rotation, SQLite persistence, and network-free HTTP-path tests |
| Project | Focus |
|---|---|
| Regularized Operator Zoo | Entropy, KL, Tsallis, and chi-squared regularized greedy operators with numerical identity checks |
| GRPO Minimal | RFT, Online RFT, and GRPO+OS on one training skeleton with a tested dependency on the operator library |
| Project | Focus |
|---|---|
| Image Captioning | VGG16 encoder, LSTM decoder, COCO loading, TPU strategy with fallback, BLEU-4 evaluation, and a CPU smoke run |
| Image Classification | Frozen VGG16 transfer learning, held-out testing, Flask inference, Docker, and synthetic end-to-end validation |
| Predictive Maintenance | Battery telemetry preprocessing, persisted feature contract, Random Forest training, evaluation, and Flask serving |
These crates demonstrate protocol mechanics in tested, in-process simulations. They are not production blockchains or deployed smart contracts.
| Project | Focus |
|---|---|
| Cross-Chain Atomic Bridge | HTLC commit and reveal, explicit message boundaries, rollback, conservation, and asset representation |
| Decentralized Voting | Commit and reveal ballots plus Merkle-root verification |
| Quadratic Voting and Liquid Democracy | Quadratic credit costs, delegation, and cycle detection |
| PoS and ZKP Voting | Stake-weighted leadership, signed votes, eligibility commitments, and a tamper-evident chain |
| PoA and ZKP Voting | Authority rotation, signed blocks, and voting protocol mechanics |
| DeFi Lending Protocol | Fixed-precision accounting, interest accrual, utilization-based rates, liquidation, and append-only events |
| Project | Focus |
|---|---|
| Sales Data ETL | SSIS package, validation and error routing, idempotent SQL setup, SQL Agent scheduling, and a portable Python reference ETL |
| SSE Coexistence Testing | Terraform-managed AWS test environment, hardened host bootstrap, and post-deployment reachability checks |
Experimental work remains under prototypes/ and is intentionally excluded from the featured project count.
The root GitHub Actions workflow runs:
- test suites for 10 Python projects;
cargo build,cargo test, and advisory Clippy checks for 6 Rust crates;terraform init -backend=falseandterraform validatefor the cloud-security configuration.
Projects that depend on external APIs, large datasets, TPU access, SQL Server, or AWS provide a network-free test path, a synthetic smoke pipeline, or a portable reference implementation where appropriate. Generated reports are excluded when values can drift between runs; commands and selected verified outputs are documented in project READMEs and RESULTS.md.
- Each project README distinguishes implemented behavior from future enhancements.
- Live OpenAI calls require user-supplied credentials; tests use deterministic or mocked paths.
- Rust blockchain projects model protocol behavior without claiming live-network security or consensus.
- Terraform is validated in CI, while deployment remains opt-in and may create billable cloud resources.
- Prototype directories are exploratory and are not represented as portfolio-complete projects.
- Secrets, local virtual environments, generated models, datasets, and runtime reports are excluded through repository ignore rules.
There is no repository-wide dependency installation because the projects use different Python, Rust, Terraform, SSIS, and cloud toolchains.
git clone https://github.com/lmdixon23/my_dev_projects.git
cd my_dev_projectsChoose a project from the index above and follow its README for installation, execution, testing, and scope.
The interactive teaching applets formerly stored here now live in lmdixon23/ai-playgrounds, with a public browser demo.
See CONTRIBUTING.md before proposing a substantial change.
Report security concerns according to SECURITY.md.
Licensed under the MIT License.