15 Technologies β Week of May 11, 2026
Open-source local inference engine purpose-built for Apple Metal GPU API. Runs DeepSeek 4 Flash at near-cloud speeds on Mac β no NVIDIA needed.
Local AI inference on MacBooks/iMacs: code generation, document analysis, agentic tasks β no data leaves the device.
Clone antirez/ds4 (GitHub 7K β). Run the benchmark on an M-series Mac. Compare tokens/sec vs cloud API. Cost: free Β· Time: 30 min
Field reps run proposal drafts, spec comparisons, and code reviews on a MacBook β no internet, no per-token costs. The whole team gets local AI on hardware they already own.
Build desktop + mobile apps from a single Zig codebase. Compiles to native binaries for macOS, Windows, Linux, iOS, Android β no Electron, no JVM.
High-performance cross-platform apps where bundle size, startup time, and memory matter. Replaces Electron for internal tools.
Clone vercel-labs/zero-native, build the demo for macOS. Measure binary size vs equivalent Electron app. Cost: free Β· Time: 1 hr
One codebase for PGC's internal tools β PM app, glass calculator, field reporting β running on Windows PCs in the office and tablets on the job site.
Unified virtual filesystem layer β AI agents see S3, Dropbox, Google Drive, GitHub, databases all as one POSIX-like filesystem. No custom integrations per source.
Giving AI agents access to scattered data: PM tools, Google Drive, project folders, email attachments β queried through a single API.
Set up mirage, connect to a local project folder + mock S3 bucket. Ask agent: "List all PDFs modified this week and summarize the glazing specs in each." Cost: free Β· Time: 2 hrs
PGC data lives everywhere β Procore, Google Drive, email, project folders. Mirage lets an AI agent query across all of it without building a custom integration per source.
Browser-based AI studio that generates, annotates, and explores 3D cellular structures. Combines AI generation with scientific visualization tooling.
Scientific visualization, material science, and via mesh-to-3D workflows β architectural and structural modeling from photos.
Upload a photo of a curtain wall panel or glass facade detail. Generate an annotated 3D model. Export to OBJ for BIM software. Cost: free (beta) Β· Time: 1 hr
Converting site photos into 3D models for client presentations and BIM integration β without a CAD operator. Could accelerate visual documentation significantly.
GPU inference engine operating near memory-bandwidth token generation speeds. Built for throughput over model size.
High-volume AI workloads: document processing, batch summarization, proposal generation β where per-token latency compounds at scale.
Run tokenspeed benchmark vs llama.cpp on same hardware. Measure throughput (tokens/sec) for a 500-token task. Calculate cost-per-token at scale. Cost: free Β· Time: 1 hr
If PGC auto-generates 50 proposals/week, a 2x inference speedup cuts AI processing costs in half. Worth evaluating as proposal volume grows.
Library of HTML slide templates designed so any coding agent produces a polished presentation deck without manual formatting. Includes charts, diagrams, layouts.
Automated report generation, AI-produced client deliverables, weekly brief publications, proposal appendices.
Prompt a coding agent: "Generate a 10-slide brief on Q2 glazing trends for a construction client using beautiful-html-templates." Cost: free Β· Time: 1 hr
Joe already generates Steve's weekly briefs this way. This library makes the output polished enough to send directly to clients β no manual formatting.
Lisp-like language in Go that boots in 7 milliseconds. Targets scripting, REPL-driven development, and embedding in Go applications.
Fast startup scripting, embedded DSLs in Go services, REPL-driven workflows where boot latency matters.
Install via go install. Run the REPL. Write a script that parses a glazing CSV and outputs a cut list summary. Compare boot time to a Go binary. Cost: free Β· Time: 1 hr
Operations team could write fast data-processing scripts without Python/JVM startup overhead. Useful for on-the-fly analysis on site.
Lisp that compiles to idiomatic Rust. Write in Lisp, get Rust output. Combines Lisp's expressiveness with Rust's safety and performance.
High-performance systems where Lisp macros and iteration primitives are valuable, but production output needs to be Rust.
Try the web playground. Write a short program using recursion and macros. Inspect the generated Rust output. Cost: free Β· Time: 30 min
Early-stage preview. But Rust-in-Lisp tooling could eventually produce high-performance native apps without requiring Rust expertise.
Cloud sandbox for AI agents with a transactional, versioned filesystem. Every action is atomic β agents can fork, branch, and roll back like git.
Safe agent experimentation, reproducible agent runs, testing AI pipelines with full rollback capability.
Run a 10-step task (e.g., "generate a material takeoff for a 5-window facade"). Inspect the version history to see every agent decision. Cost: free tier Β· Time: 1 hr
Before deploying an agent for submittal reviews or proposal generation, test it against historical projects with full rollback β catching errors before they affect live workflows.
Open-source multi-agent PR review system for Claude Code. Multiple specialized agents (security, style, logic) debate and synthesize findings.
Automated code review in CI/CD β multiple quality dimensions checked simultaneously by specialized agents.
Install adamsreview on a sample PR. Configure to review a Python script. Compare findings to a single-agent review. Cost: free Β· Time: 1 hr
PGC's internal tools codebase gets automated multi-perspective reviews on every commit β security, style, logic errors β without a dedicated reviewer.
Version control system with git-like semantics designed for AI agents. Tracks agent state, enables branching, merging, and diff viewing of agent "commits."
Agent audit trails, reproducible agent behavior, team workflows where multiple agents collaborate with version control.
Run re_gent alongside a coding agent task. After each agent action, inspect the diff. Reset to a previous checkpoint and compare outcomes. Cost: free Β· Time: 1 hr
If PGC runs AI agents for proposals or code generation, re_gent gives an audit trail of every decision β important for quality control and client accountability.
Bun-native browser automation library. Mimics real user interactions with full DOM fidelity β not just Playwright-style screenshots.
Automated procurement flows, competitor monitoring, permit filing, legacy portal automation where visual verification matters.
Install mochi.js. Task it to navigate a glass supplier portal, log in, find Β½" tempered clear glass pricing, return a structured table. Cost: free Β· Time: 2 hrs
Automates the repetitive online quoting workflow β checking supplier pricing, logging into portals β without brittleness of older automation tools.
Airbyte's agent framework gives AI agents contextual awareness across 300+ data connectors β databases, SaaS tools, data warehouses. Agents query live data without custom integrations.
Cross-source analytics, AI-assisted business intelligence, agents that answer questions against live data from multiple systems.
Connect Airbyte to a mock Postgres database of PGC projects + Google Sheet of pricing data. Ask: "Which projects this quarter used >50% tempered glass and what's the total margin?" Cost: free tier Β· Time: 2 hrs
A single AI query across Procore, pricing sheets, and accounting β without a data engineer to build the pipeline first.
AI generating CAD models and fabrication drawings from text prompts. Open-source (DeepCAD) and commercial tools (Autodesk Forma, nTop) advancing rapidly.
Custom flashing design, curtain wall component iteration, fabrication-ready drawings from sketches, automated bill of materials from descriptions.
Try DeepCAD (open source) with: "Aluminum glazing anchor bracket, 4in x 6in, with β in mounting slots, hot-dip galvanized." Export to DXF. Cost: free Β· Time: 1 hr
Custom connection details and fabricator drawings could be prototyped in-house in hours instead of waiting weeks for shop drawings β a major RFP response advantage.
Compact, code-compliant balcony solar mounting systems that clip onto railings without penetration. Growing 3x YoY in Europe; now entering US after recent NEC code updates.
Renters, condo owners, apartment dwellers who can't install rooftop solar. $500β$2,000 alternative offsetting 30β50% of electricity use.
Research top 3β5 balcony solar brands (e.g., Sunrun, SunPower Balcony). Identify if any are actively recruiting glazing/construction partners for installation. Cost: free Β· Time: 1 hr
Glazing companies already do exterior building envelope work. Balcony solar is a natural adjacent service β same crews, same access, new revenue. US market about to explode.
| # | Technology | Category | MVP Cost | Time |
|---|---|---|---|---|
| 1 | DeepSeek 4 Flash Local Inference | AI/ML Β· Edge | Free | 30 min |
| 2 | Zig-Native Cross-Platform | Dev Tools | Free | 1 hr |
| 3 | Agent Virtual Filesystems | AI/Infra | Free | 2 hrs |
| 4 | AI 3D Cell Generation | AI/3D | Free | 1 hr |
| 5 | Speed-of-Light Inference | AI/Infra | Free | 1 hr |
| 6 | HTML Templates for Agents | Dev Tools | Free | 1 hr |
| 7 | Clojure-in-Go | Languages | Free | 1 hr |
| 8 | Rust-in-Lisp | Languages | Free | 30 min |
| 9 | Versioned Agent Sandboxes | AI/Infra | Free | 1 hr |
| 10 | Multi-Agent PR Review | AI/DevTools | Free | 1 hr |
| 11 | Git for AI Agents | AI/Infra | Free | 1 hr |
| 12 | Bun Browser Automation | Automation | Free | 2 hrs |
| 13 | Cross-Source Agent Context | AI/Data | Free | 2 hrs |
| 14 | Text-to-CAD | AI/Design | Free | 1 hr |
| 15 | Balcony Solar Boom | Clean Tech | Free | 1 hr |