# Paul Edward Harris 4th

> Paul Edward Harris 4th (also written Paul E. Harris IV) is an independent AI/ML learner and researcher from Connecticut, studying on Cape Cod, Massachusetts, US. He learns machine learning by building real projects with AI tools as a development partner, and documents results honestly, including negative ones.

His projects are built with heavy AI assistance; he directs, runs, debugs, and verifies the work, and is actively learning to understand each part himself.

## How he thinks

- Ownership: he owns his path completely; he didn't come up through the usual route and doesn't let that limit what he builds. "Someone who refuses to stay broken doesn't fit a neat label."
- Chaos, not clockwork: small inputs can shift big outcomes, as in the Lorenz system. This drives his research on forecasting chaotic systems (SynechismCore).
- Build in public: he builds end to end, tests, and shares real results including what didn't work, with AI as his development partner.
- Verify before believing: he treats claims from media, institutions, or his own models as things to check, not accept. This is the idea behind Narrative Radar, Firm Radar and his podcast, Listening Between the Lies.

## Projects

- [SynechismCore](https://github.com/seek-k-nowledge/SynechismCore): Independent ML research. Technical report: "Continuous-Time Latent Dynamics for Long-Horizon Chaotic Forecasting under Irregular Sampling" (Paul E. Harris IV, Independent Researcher, Mashantucket Pequot Nation; SynechismCore Technical Report v1.1.0, November 2026, 15 pages; DOI [10.5281/zenodo.23772387](https://doi.org/10.5281/zenodo.23772387); [PDF](https://github.com/seek-k-nowledge/SynechismCore/blob/master/paper/Harris_2026_Continuous-Time_Latent_Dynamics_Irregular_Sampling.pdf)). Introduces SynechismV23, a forecasting architecture he designed: a continuous-time latent neural ODE with an elastic attractor term, golden-ratio sub-step scheduling and a laminar bypass. Asks whether it keeps chaotic forecasts accurate longer than Transformer, LSTM and simplified Mamba-style forecasters when observations arrive at uneven times and all models get the same timing information. Lorenz-63 under four sampling regimes (regular, jitter, half missing, exponential) with the same nominal average rate; all models trained for 200 epochs; 5 seeds; scored by valid prediction time (VPT) in Lyapunov times. SynechismV23 had the longest VPT in all four regimes; under exponential sampling: SynechismV23 2.03, LSTM 1.78, Transformer 0.55, Mamba-style 0.50. Transformer and Mamba-style baselines lost about 80% of their VPT from regular to exponential sampling; the lead over the LSTM under irregular sampling is small (4 of 5 seeds, 9–14%) and not yet statistically significant at the seed level, and the LSTM degrades least as timing becomes uneven. Limits: Lorenz-63 only, one model trained per regime, baseline decoding differs (block output vs step-by-step), simplified Mamba-style baseline, ablations at 50 epochs only, not peer reviewed. Next: stronger baselines, a Kuramoto–Sivashinsky turbulence benchmark, real robotics data, then arXiv and a 2027 workshop submission. Code and results are MIT-licensed. Project history: SynechismCore began as an exploratory study of forecasting under periodic correction; that study was discontinued and is superseded by this report. Only the irregular-sampling results are current claims.
- [GraphRAG Agent](https://github.com/seek-k-nowledge/enterprise-graphrag-agent): Working prototype (v0.x), actively developed toward production. Self-grounding knowledge graph pipeline with cited answers. Roadmap: authentication, security hardening, claim typing, load testing. (Python, FastAPI, LangGraph, Neo4j, Streamlit)
- [Firm Radar](https://github.com/seek-k-nowledge/firm-radar): Working prototype (v0.1). A private narrative-intelligence operations room for each client: a threat radar, a crowd check that flags possible coordinated inauthentic behavior for an analyst to review (accuracy not yet measured), network and spread maps, and weekly briefs with confidence levels. Python, runs locally, public sources only.
- [Narrative Radar](https://github.com/seek-k-nowledge/narrative-radar): Working prototype (v0.1). A real-time news monitor that flags when a topic suddenly surges, across news, social, viral trends and video. Built to research his podcast, Listening Between the Lies.

## Profiles

- [Portfolio](https://seek-k-nowledge.github.io/)
- [ORCID 0009-0009-3285-4610](https://orcid.org/0009-0009-3285-4610)
- [Zenodo: technical report, DOI 10.5281/zenodo.23772387](https://doi.org/10.5281/zenodo.23772387)
- [GitHub: seek-k-nowledge](https://github.com/seek-k-nowledge)
- [LinkedIn](https://www.linkedin.com/in/paul-edward-harris-4th-3a689941b)
- [X: @PaulEHarris4th](https://x.com/PaulEHarris4th)
- [Hugging Face: seek-k-nowledge](https://huggingface.co/seek-k-nowledge)
- [Facebook](https://www.facebook.com/PaulEdwardHarris4th/)
- [Facebook Page](https://www.facebook.com/PaulEdwardHarrisIV/)