A curated map of AI agents, robotics/world-model experiments, RAG systems, data pipelines, Telegram/TON automation, and developer tools built across public and private repositories.
Go, local LLMs, CLI automation
A shell interceptor and local-AI command agent that catches mistyped commands, suggests corrected terminal actions, and can operate in agent mode from the CLI.
What it proves
Turned an everyday terminal failure mode into an assistive workflow, combining command parsing, local model prompting, and executable automation loops.
Python, Faster Whisper, AppleScript, vector search
A local macOS assistant that combines voice input, document retrieval, graph-style context, and AppleScript control.
What it proves
Explored a privacy-preserving desktop assistant architecture where local context and OS automation are first-class capabilities.
Go, macOS, launchd, Docker, SSH
A native macOS operations HUD for scripts, launchd agents, Docker containers, SSH services, and remote jobs.
What it proves
Condensed scattered local and remote runtime state into one practical control surface for daily engineering operations.
Python, static analysis, local-first tooling
A local-first repository scanner that converts codebases into structured summaries while keeping source private.
What it proves
Created a repeatable way to understand unfamiliar repositories without uploading proprietary code to external services.
Python, code indexing, knowledge catalogs
A repository-to-knowledge-catalog tool for compact codebase understanding and fast project orientation.
What it proves
Focused on making source trees searchable, summarizeable, and useful as downstream context for humans or AI agents.
Python, V-JEPA, robotics, planning
A Mac-first robotics scaffold for V-JEPA/V-JEPA2 latent world models, action-conditioned dynamics, planning, and SO-101 adaptation.
What it proves
Connected representation learning ideas to concrete robot-data workflows and planning experiments.
A shell interceptor and local-AI command agent that catches mistyped commands, suggests corrected terminal actions, and can operate in agent mode from the CLI.
Turned an everyday terminal failure mode into an assistive workflow, combining command parsing, local model prompting, and executable automation loops.
A local macOS assistant that combines voice input, document retrieval, graph-style context, and AppleScript control.
Explored a privacy-preserving desktop assistant architecture where local context and OS automation are first-class capabilities.
A native macOS operations HUD for scripts, launchd agents, Docker containers, SSH services, and remote jobs.
Condensed scattered local and remote runtime state into one practical control surface for daily engineering operations.
A local-first repository scanner that converts codebases into structured summaries while keeping source private.
Created a repeatable way to understand unfamiliar repositories without uploading proprietary code to external services.
A repository-to-knowledge-catalog tool for compact codebase understanding and fast project orientation.
Focused on making source trees searchable, summarizeable, and useful as downstream context for humans or AI agents.
A Mac-first robotics scaffold for V-JEPA/V-JEPA2 latent world models, action-conditioned dynamics, planning, and SO-101 adaptation.
Connected representation learning ideas to concrete robot-data workflows and planning experiments.
An active-inference framework for backprop-free generative world modeling, robotics benchmarking, and smolVLA/LeRobot experiments.
Built a testbed for comparing predictive world-model behavior against embodied control needs.
Aegis Certification Protocol platform for benchmarking robotic policies across safety, mechanical integrity, and mission success.
Framed robotics evaluation as a certification problem with explicit pass/fail evidence instead of informal demos.
A POV video quality pipeline that detects hand presence, stability, lighting, and validity for robot-training data.
Moved dataset quality checks earlier in the robot learning pipeline so weak trajectories can be filtered before training.
A unified multimodal RAG server combining LightRAG with ingestion, indexing, web access, and API workflows.
Turned retrieval from a notebook technique into a service layer that can support applications and user-facing flows.
A personal AI doppelganger experiment trained around Telegram history for reflection and conversational simulation.
Explored how personal archives can become interactive memory systems while raising questions about identity, consent, and recall.
A Telegram betting and binary prediction-market bot with TON blockchain payments and dynamic market workflows.
Combined bot UX, payment rails, market-state management, and operational automation into a single product loop.
A Go upload backend using resumable upload hooks and Grably configuration models.
Built reliable large-file intake primitives for media-heavy data operations.
A dashboard and upload worker for browsing, previewing, signing, and tracking large file uploads.
Made storage operations inspectable and supportable for nontrivial datasets and transfer workflows.
Data exporter and dataset tooling for project configurations, URL retrieval, summaries, and command-driven exports.
Created operational glue between raw collection systems and usable dataset artifacts.
A local interview copilot that transcribes conversations and generates real-time suggestions in a web UI.
Tested a low-latency human-assistance loop where speech, context, and generated suggestions must stay synchronized.
A hybrid steganography engine for robust digital watermarking across images, audio, and text.
Explored resilient information hiding as a practical response to provenance, watermarking, and media-authenticity problems.
A cellular-automata physics engine simulating elemental interactions, heat transfer, and bending-style abilities.
Implemented a rule-based world where simple local interactions produce richer emergent dynamics.
A tiny hand-crafted decoder-only transformer implementation inspired by Theia Vogel's explainer.
Used implementation from first principles to make attention, residual streams, and token prediction mechanics concrete.
A managed workspace for writing, organizing, and archiving scientific papers.
Shaped paper-writing as an operational workflow with structure around drafts, project state, and archived artifacts.