Hermes Wiki
TechResearch/_index

TechResearch

Recurring personal knowledge system to stay aware of the latest tools, frameworks, models, papers, and industry shifts — and to deeply research specific technologies, problems, or innovations on demand. Pure signal, zero fluff.

Rules for this folder

  • Existing research documents are read-only — never remove or overwrite them.
  • New research can always be added as a new subfolder or file.
  • Each topic gets its own subfolder with a research-<topic>.md as the primary deep-dive and any companion guides alongside it.
  • system_prompt — folder-level behavior rules (triggers, output structure, processing rules); the original, do not edit without explicit instruction

Context Compression & Token Efficiency

  • research-headroom — Headroom: context compression layer for AI agents (60–95% token reduction, reversible, local-first)
  • headroom-compression-guide — Headroom compression levels, semantic risk, enterprise deployment modes, real experiment results
  • research-rtk — RTK (Rust Token Killer): shell output compression proxy for agentic coding sessions
  • rtk-vs-headroom — Side-by-side comparison: where each tool compresses, decision matrix, real benchmarks

Voice / Audio AI

  • chatterbox_experiment — Chatterbox (Resemble AI): open-source zero-shot voice cloning TTS, MIT-licensed successor to VoiceBox/XTTS-v2; includes hands-on paralinguistic tag notes
  • chatterbox_implementation_log — Blow-by-blow build log: self-hosting Chatterbox Turbo on Apple Silicon (MPS) via devnen/Chatterbox-TTS-Server, dependency conflicts and fixes, frontend integration
  • chatterbox_implementation_guide — Synthesis of research vs. reality: what the research called correctly, what only surfaced on real hardware, full dependency archaeology, and the shipped integration

Cross-Project Infrastructure (Auth & Payments)

Reusable setup knowledge for infrastructure every self-built project needs — not research on a single tool, so these don't follow the research-<topic>.md pattern above; they're living reference docs revisited each time a new project comes online.

  • clerk_based_auth — Combined auth architecture across FullStackFusions' projects (wiki, Courses, Localz): project map, provider choices, one-Clerk-app-per-project reasoning
  • clerk_production_setup_walkthrough — Actual dashboard walkthrough for setting up a Clerk production instance (domain gotchas, primary vs. secondary application)
  • clerk_free_tier_access_control — Restricting sign-in to specific people on Clerk's free tier via app-code authorization instead of the paid allowlist feature
  • stripe_based_payments — Stripe setup notes: account-per-project reasoning, pricing, env var conventions, local CLI testing, reusable Checkout+webhook integration shape. First used for the Courses project

Tech Concept Explanatory Animations

Building interactive, code-driven animations to explain tech concepts — for course content, a portfolio site, or a YouTube channel.

  • what_I_can_achieve — Brainstorm: what's buildable with React Flow + Framer Motion + Tailwind + HTML/CSS/JS — network/architecture visualizations, data structure & algorithm walkthroughs, system design mechanics, code execution/interpreter animations
  • code_based_animation_setup — Three-tier setup reference (escalate only when the previous tier can't do the job): plain HTML/CSS/SVG/JS → React Flow + Framer Motion → bridging Excalidraw's hand-drawn feel (Excalifont, rough.js) into code-driven diagrams
  • excalidraw — Excalidraw + excalidraw-animate setup reference: static architecture diagrams and "watch it get drawn" reveals; scene JSON generation, stroke-order animation, icon sourcing

Harness Engineering

  • ecc_agent_harness_toolkit — ECC (affaan-m/ECC): MIT-licensed, 239K-star agents/skills/rules/hooks/memory toolkit for Claude Code and other harnesses; gated TDD workflow, Unified Memory Vault, AgentShield security scanning, token-optimization settings
  • agent_harness_hands_on — Three hands-on experiments (harness ladder benchmark, multi-agent coding pipeline, deep research agent) measuring how harness infrastructure — not model capability — determines agent reliability
  • harness-rtk-headroom-synthesis — Three-layer synthesis: Agent Harness (reliability), RTK (shell compression), Headroom (context compression) — how they complement rather than compete
  • agent_harness_engineering — unprocessed bookmark, needs a research pass
  • harness_engineering — unprocessed bookmark, needs a research pass
  • harness_design_long_running_app — unprocessed bookmark, needs a research pass
  • the_anatomy_of_an_agent_harness — unprocessed bookmark, needs a research pass
  • deepagents_vs_goose — unprocessed bookmark, needs a research pass
  • llm_wiki — unprocessed bookmark, needs a research pass
  • Superpowers (obra) — cross-harness agentic-coding skills framework (~277K stars); full entry under "Personal Branding & Career Positioning" below since that's the angle it was researched for

Personal Branding & Career Positioning

Case studies and patterns for marketing/positioning yourself — interviews, personal brand, content creation. Started 2026-08-25.

Networking

Agent Skills

  • research_agent_skills_setup — Agent Skills (SKILL.md) format: creating, enabling, and distributing across Claude Code, the open agentskills.io spec, and cross-agent tooling (npx skills, officialskills.sh); includes a Skills-over-MCP hosting angle relevant to Hermes

Lightning Talks

  • Lightning Talk — October 2026 — RBC internal lightning-talk prep (15 min + 15 min Q&A): event requirements, three themes, 10 tracks (pick 3), depth levels, and the chosen 3-talk lineup
  • lightning-talk-oct2026-requirements — Event constraints, theme/track/depth mapping, and the full 3-talk lineup with abstracts
  • talking-to-the-network-source-of-truth — Flagship talk "The Network That Answers Back": agentic NetOps over MCP against a NetBox/Nautobot source of truth, approval-gated + digital-twin sandboxed — abstract, PoC architecture, demo run-of-show, Q&A prep

Agent Products

Hermes Wiki