Featured work
Each asset is a complete, working system — designed, built, tested, and documented solo. Serial numbers are real: this is an inventory, not a mood board. Click through for the full build story: the problem, the architecture, and the measured results.
Constellation COO
Patent-pending Constraint-Optimized Orchestration for LLM agent lifecycles. Multi-dimensional fitness scoring routes, spawns, and retires agents against real cost, latency, and budget constraints.
- Stack
- Python · pytest
- Scale
- 13 modules · 39 tests
- Benchmarks
- +16.5% SLA · −41% load variance
- Status
- Patent pending
TaskFlow
Three-phase AI agent — Plan, Execute, Verify — with markdown-backed persistent memory, a typed LangGraph orchestration graph, and full deployment documentation for Vercel, Docker, and VPS.
- Stack
- Next.js 15 · React 19 · LangGraph
- Scale
- 30 files · 36K words of docs
- Status
- Complete · Documented
Scanline
Phone-scan-to-searchable-library for physical photographs. Perspective correction, face clustering, duplicate detection by perceptual hash, full-text search, and AI vision tagging.
- Stack
- React · OpenCV.js · face-api.js
- Scale
- 2,300+ line single-file build
- AI
- Claude Vision wired end-to-end
- Status
- MVP complete
HostGraph
Margin-leak detection command center for restaurant groups: vendor analysis, reorder logic, shrinkage tracking, and alerting across six dashboard views. Clean TypeScript build, verified end to end.
- Stack
- React 19 · Vite · Tailwind
- Scope
- 6 views · full margin engine
- Status
- Demo-ready
StoreForge
AI dropshipping orchestrator: a six-stage spawn pipeline generates market research and brand identity via live LLM calls, backed by Prisma, Stripe webhooks, SSE streaming, and subdomain routing.
- Stack
- Next.js · Prisma · Stripe
- Scope
- Full backend + storefront
- Status
- Acquisition asset
Autoblog Pipeline
Fully automated publishing system: six source fetchers, composite scoring and dedup, LLM generation with schema-validated MDX, and GitHub-commit deploys to Cloudflare Pages. Runs for $0/month.
- Stack
- Next.js · TinaCMS · Groq
- Sources
- Reddit · HN · DEV · RSS · YT · Brave
- Status
- Production pipeline
VERITAS
A benchmarking harness that ranks frontier AI models on hidden-test pass rate across a full suite — leaderboard, coverage matrix, and per-task explorer.
- Metric
- Hidden-test pass rate
- Views
- Leaderboard · matrix · tasks
- Status
- ● Live & explorable
CipherHorizon
Scans certificates, dependency manifests, and live domains for cryptography that won't survive a quantum computer — scoring dashboard and continuous monitoring.
- Inputs
- Certs · deps · domains
- Model
- $149 report · $99/mo
- Status
- ● Live product
What I work in
Languages & Frameworks
- Python — engines, pipelines, pytest
- TypeScript / JavaScript
- React 19 · Next.js 15
- Tailwind · Vite
AI & Agents
- LLM orchestration — routing, lifecycle, cost control
- LangGraph agent workflows
- Anthropic & OpenAI APIs in production
- Vision pipelines — OCR, tagging, clustering
Systems & Infra
- Prisma · Stripe billing & webhooks
- GitHub Actions CI/CD & cron
- Vercel · Cloudflare deploys
- SSE streaming · OAuth2
Beyond Code
- Patent prosecution — prior art, claims drafting
- Benchmarking with statistical rigor
- Technical writing — docs people can actually use
- Product strategy — build-to-revenue, build-to-sell
How an asset gets built
There's no team to hand work off to, so every asset follows the same disciplined loop — the same one a founding engineer would run inside a company, minus the meetings.
Diagnose the constraint
Every build starts by naming the real bottleneck — a cost ceiling, a latency SLA, a margin leak — instead of the feature that's easiest to build. Constellation COO exists because agent fleets were being scheduled on intuition, not math.
Design against evidence
Architecture decisions get benchmarked, not assumed. Fitness-scoring weights, retry budgets, and cache strategies are tuned with bootstrap confidence intervals and load tests before they ship, not after a user complains.
Build for someone else to run it
Solo doesn't mean sloppy. Every asset ships with deployment docs, typed interfaces, and a test suite — TaskFlow shipped with 36K words of documentation because the goal is systems a stranger could operate, not just demo.
Protect the IP that matters
When an algorithm is genuinely novel — like Constellation's constraint-optimized routing — it gets drafted into formal patent claims with prior-art analysis, not just shipped and hoped for.
Package for the next owner
Every asset is built as if due diligence starts tomorrow: clean commit history, documented architecture, and a data room's worth of context — because a system that can't be handed off isn't finished.
The builder
I'm Nic Albertson — a founder-engineer in Fall River, Massachusetts who treats software like real estate: find undervalued ground, build something solid on it, and make it produce.
My flagship work is Constellation COO, a patent-pending algorithm for managing fleets of AI agents against hard cost and latency constraints — drafted into seven claims, benchmarked with bootstrap confidence intervals, and implemented as a tested Python engine.
Around it sits a catalog of shipped products: agent frameworks, commerce platforms, automation pipelines, and vertical SaaS tools. I design, build, test, document, and position every one of them myself — from first commit to data room.
When I'm not shipping, I write about the engineering and business decisions behind these builds — see the insights section for deeper dives on orchestration design, patenting as an indie engineer, and what it actually takes to run 20+ products solo.
Recent writing
Notes from the build log — on orchestration design, IP strategy for indie engineers, and what it takes to run a product catalog solo.
Constraint-Optimized Orchestration: Rethinking How Agent Fleets Get Scheduled
Why fitness-scoring beats round-robin and greedy heuristics when you're routing LLM agents against real cost and latency budgets.
Why I Patent My Own Algorithms as a Solo Engineer
Filing a provisional as an indie builder felt like overkill — until I understood what it actually protects, and what it costs to skip.
Shipping 20+ Micro-SaaS Products Solo: A Playbook
The repeatable loop — scope, build, instrument, decide — that makes a one-person catalog possible without burning out.
Hire the builder, or own the asset
Everything above is for sale in one form or another: fixed-scope engagements, source licenses, or full acquisition. Fixed quotes, no hourly billing, no subcontracting — the person you email is the person who builds it.
Systems & AI-Spend Audit
A week inside your AI pipeline or full-stack system — findings, benchmarks, and a prioritized fix list you can act on immediately.
Details →Build Sprint
A scoped system shipped to your infrastructure with tests, typed interfaces, and docs a stranger could operate from.
Details →Take an asset home
Any serial number in the index is available as an annual source license — or as a full acquisition with IP assignment and a proper data room.
Licensing & acquisition →Looking to acquire, hire, or build something together?
Pick the lane that fits — each one lands in my inbox with the context I need to reply with something concrete, within two business days.
Book an engagement
An audit, a build sprint, or fractional founding-engineer work — fixed quotes only.
Start the email →Request a data room
Architecture docs, test suites, commit history, and benchmark methodology — under NDA if needed.
Start the email →License an asset
Run any catalog asset inside your organization, with docs and 30 days of deployment support.
Start the email →