What we have built,
and what we learned.

An AI phone concierge that turns inbound calls into CRM leads

A voice agent that answers inbound seller calls, qualifies them on a fixed template and writes the lead straight into the CRM.

Built against a written acceptance test: inbound call, AI conversation, populated CRM record

An AI trip planner that builds day‑by‑day itineraries from real places

A travel agent’s planning tool: answer a few questions and get a streamed, editable itinerary grounded in real place search, saved and shareable.

Working end-to-end product, from intake to shareable trip page

Cutting a WordPress site from 59 stylesheets to 1

A performance audit and clean-up for a design and fabrication company’s marketing site, with a migration plan to Next.js on Vercel.

On staging, CSS files went from 59 to 1 and requests from 142 to 72

A live voice co‑pilot that coaches sales reps during calls

A prototype for an equipment-rental enterprise: it listens to a call, suggests what to say next and writes the summary afterwards.

Demonstrated on seeded data; not run against live customer calls

A faster marketing site for a voice‑AI agency

A rebuild of an agency’s public website on Next.js, with an interactive product simulator and a demo-request flow.

A voice agent that browses the web for people who cannot use a screen

Speak a task, and an agent drives a real browser and talks the results back. Built with safety limits and an audit log for caregivers.

Working on localhost; not deployed

One platform, many client voice agents, no per‑client code

The multi-tenant foundation we start client builds from: onboard a client with configuration, and swap vendors without touching the core.

167 automated tests passing

A client‑lifecycle system from first lead to handover

An internal system for running an agency’s client journey: lead, proposal, milestones, invoices and handover, with two approval gates.

84 automated tests passing across the rules engine and the platform layer

A pricing engine that turns vendor prices into margins

A small library that prices AI products from incompatible vendor rates and shows what a business model earns, month by month.

17 automated tests passing

Open‑source memory and guardrails for the coding agent you already use

A free command-line tool that gives Claude Code, Codex and other coding agents durable knowledge, memory and guardrails as plain markdown.

Built in public since June 2026, with over a thousand commits

A research engine that turns sources into cited, validated articles

The pipeline behind our published dispatches: sources, notes, validation and publishing, with checks at every step.

The checking tool has 74 passing tests

How can a self-learner reach MS-level CS theory?

The curriculum, sequence and resources for MS-level CS theory, and how to know you got there.

How can a self-learner build professional software-engineering craft?

Mid-level engineering craft, free resources, professional shipping, and learning with coding agents.

What is an independent researcher, and how do they sustain themselves?

What defines independent researchers, what keeps them going, and how they trade off against academia.

Which APIs and tools query scientific papers programmatically?

How researchers and agents retrieve papers, citation graphs and author metrics through APIs.

How is the R&D-as-a-service market structured?

The segments, players and value propositions of R&D as a service.

How do physical phenomena give rise to formal mathematical systems?

From physics-informed ML to nature-inspired mathematics: how observation and formal systems relate.

How does someone become a sustainable independent researcher?

The steps, finances, tools and credibility signals needed to research independently for the long term.

How does peer review work at top ML conferences?

The review pipeline, anonymity rules and yearly timeline of ICML, NeurIPS, ICLR and CVPR.

How do aneural biological systems process information and learn?

How plants, slime moulds and single cells sense, decide, remember and communicate without a brain.

How should a beginner learn programming without tutorial hell?

Why novices stall on syntax, and how a mental model of the machine gets them building.

Do Product Hunt launches still measure real startups?

Whether launch counts still track startups being built, now that agents make shipping cheap.

Where is the moat once foundation models commoditise?

If models become interchangeable, whether the desktop agent harness becomes the defensible layer.

A tech digest that ranks stories by real human effort

A daily newspaper-style digest that only published stories after they survived a settling delay, scored by a “Human Pain Index”.

Retired on purpose; the fetchers are being reused in the research engine

A timetable engine that catches clashes as you build

A scheduling engine for schools that detects teacher, room and group conflicts in real time on flexible time slots.

Shelved after the first phase; the scheduling core has automated tests

An engine that compiles STEM problems into step‑by‑step lessons

A pipeline that turns a problem blueprint into a verified, step-by-step visual explanation, ready to present.

Kept as an experiment; test files exist for the compiler, parser and linter

Three startups from idea to concept MVP on a 30-day turnaround

Product work at a venture studio: taking idea-stage founders to a testable concept, and teaching the method.

3 startups taken from idea to concept MVP on a 30-day turnaround