COUTO PINTOS MANUEL

COUTO PINTOS MANUEL

@manuelcoutopintos

Universidade de Santiago de Compostela En ITH desde 6/8/2026
rustsveltepytorchnlpsupabasepostgresscloudflare

Manuel Couto Pintos

profile views

💻 Software Engineer · PhD (cum laude)

I'm a software engineer who did a PhD (cum laude). I design, build, and ship complete systems end-to-end — from distributed Rust runtimes and P2P apps to AI tooling and interactive web interfaces. I like owning things from architecture to deployment, and keeping them running. My doctoral research was in ML / NLP.


🛠️ Tech Stack & Tools


🚀 Featured Projects

⚙️ AI & Research Infrastructure

Pipeline-and-filter framework with hash-based caching for reproducible, modular ML experiments. Compute once, reuse everywhere — across runs and across teams. Published in SoftwareX (2026).

PythonCachingReproducibility

Website →Paper →

The next step beyond LabChain: a computational-graph runtime in Rust (+ PyO3). Declare a graph as one expression, run it across machines, cache what is worth keeping, train it and search over it — then ask it what it did: the run drawn, the health of a network diagnosed from the stored record, an architecture opened up, and what an edit invalidated before paying to find out. You think it, Soma somatizes it.

RustPyO3DistributedObservability

Docs →Repo →PyPI →

An exploratory temporal vector database in Rust that treats time as a native geometric dimension of embedding space — storing vectors as trajectories, not snapshots (velocity, drift, causal retrieval).

Rustcrates.ioPyPIHNSW

Docs →early research prototype

An experimentation environment for agentic systems: define complex agentic flows, then execute and compare them in parallel. Multi-model comparison and a visual Agent Studio, on a Rust + SvelteKit stack. Grew out of CHATTY, a tool to build and evaluate customizable LLM assistants.

SvelteKitRust (Axum)pgvector

chatty-lab.com →prototype

🌐 Decentralized & Social

A decentralized (P2P) social network that runs as a node on your own device — no ads, no manipulative algorithm, local-first and ephemeral by default. Semantic discovery computed on-device. Made in Galicia 🌿.

Rustlibp2pTauri 2Svelte 5WebRTC

Repo →Landing →alpha

🛰️ ITH

The showcase for Galician university talent — a platform where students from USC, UDC and UVigo publish their apps, games and inventions, find testers and attract partners, collaborators and investors.

SvelteKitSupabaseCloudflareRust

Private · in development

🧠 Research — Neuromorphic NLP

⚡ Mixture-of-Symptoms (MoS)

Neuromorphic symptom routing (LIF spikes / CfC) for early depression detection on social media, with per-user, per-symptom gating anchored to BDI-II for interpretability.

PyTorchSNN/LIFCfC

Private · under review

📈 When Topics Spike

A neuromorphic topic model for time-resolved theme evolution in early-risk text collections, bringing spiking dynamics to temporal topic modeling.

PyTorchSpiking NNTemporal

Private · under review


📝 Selected publications
  1. Exploiting topic analysis models to explore psychological dimensions in social media dataScientific Reports (2026, Q1). Read →
  2. A study of word embedding models for measuring topic coherenceKnowledge and Information Systems (2026, Q2). Read →
  3. LabChain: Enabling reproducible and modular scientific experiments in PythonSoftwareX (2026). Read →
  4. Analyzing Gambling Addictions: A Spanish Corpus for Understanding Pathological BehaviorFindings of the ACL: EMNLP 2025. Read →
  5. Temporal Word Embeddings for Early Detection of Psychological Disorders on Social MediaJournal of Healthcare Informatics Research (2025, Q1). Read →
  6. Comparison of Clustering Algorithms for Knowledge Discovery in Social Media PublicationsSEPLN / Procesamiento del Lenguaje Natural (2024). Read →
  7. Temporal Word Embeddings for Early Detection of Signs of DepressionCIRCLE 2022 (CEUR Vol-3178). Read →

Full list on Google Scholar.


📈 GitHub Activity

github contribution snake

Repos de GitHub

Proxectos

Colaboracións

Aínda sen colaboracións rexistradas.