I ship the product surface and the reliability layers behind it: private AI workspaces, autonomous coding agents, memory, verification, tool-call recovery, document intelligence, and business automation.
The value is not “I made a lot of repos.” The value is that the same failure points show up in every serious AI product: privacy, memory, tool reliability, source trust, and useful UX.
HammerLock is framed around founders, lawyers, analysts, and operators with sensitive work. Craig is framed around builders who need a reviewable agent loop. The projects are packaged as user workflows, not just experiments.
Tool Use Guardian, Prompt Condenser, Real-Time Verifier, RecallMax, and Surgical Code Editor show the infrastructure required to make agents cheaper, safer, more observable, and easier to trust.
The portfolio supports two lanes: startup AI engineering roles and practical client offers like local-business sites, lead capture, CRM, follow-up, and approval-first social content.
I started with a simple problem: AI is powerful, but most AI tools are too cloud-dependent, too forgetful, too unreliable, or too abstract to trust with real work.
So I built a lot of real pieces: some are products, some are libraries, some are experiments, and some are demos. The impressive part is not pretending every repo belongs in one perfect diagram. It is the range: privacy, agents, code, documents, reliability, business software, and go-to-market thinking.
The portfolio is organized by what each project proves. Where projects naturally work together, I show the workflow. Where they stand alone, I let them stand as shipped work.
Private AI assistant with encrypted memory, vault storage, and specialist agents.
Autonomous software engineer that inspects workspaces, edits files, and runs commands.
RecallMax, Guardian, Verifier, Condenser, Editor, and Reviewer solve common agent failure points.
Local RAG and codebase intelligence give agents grounded private knowledge.
Business dashboards and marketplace experiments show range beyond one product category.
One request can move through the whole system: research, verify, compress, remember, draft, check, and store the result privately.
User:
Research competitors and draft a pitch deck.
HammerLock:
Routing to Strategist agent...
Web Scraper:
Collected competitor pricing, product pages, and public claims.
Real-Time Verifier:
Checked source URLs and structured data.
Prompt Condenser:
Reduced research context while preserving URLs and facts.
RecallMax:
Loaded company positioning and prior decisions.
Tool Use Guardian:
Wrapped calls with retries, timeouts, and JSON repair.
HammerLock:
Draft saved to encrypted vault.The user gives a real outcome: research competitors, draft a deck, inspect a repo, or build a feature.
Local docs, repo structure, prior memory, user preferences, and project history stay under user control.
Retries, timeouts, repair metadata, validation checks, and source trust scores make the workflow less fragile.
The output is saved where the user can reuse it, audit it, and feed it into the next task.
Recruiters and customers do not run code first. They look for a fast proof moment: what does this do, who uses it, and why does it matter?
For founders, lawyers, finance teams, and operators who need AI help without leaking sensitive strategy, contracts, or documents.
Task:
Research three competitors and draft a pitch-deck outline.
Agent:
Strategist selected.
Sources collected: 12
Verified URLs: 12/12
Memory loaded: positioning, ICP, prior notes
Output:
1. Market wedge
2. Competitor table
3. Differentiation slide
4. GTM narrative
5. Risks and proof pointsFor founders, solo builders, and teams that need fast implementation proof: repo context, a plan, targeted edits, checks, and a readable diff.
User:
Add PDF export to the report page.
Craig:
Scanning app routes...
Found report component.
Editing export action.
Adding download button.
Running build...
Result:
3 files changed
Build passed
Next: verify generated PDF layoutFor production AI teams: fewer silent failures, cheaper context, repaired outputs, verified links, and clearer failure metadata.
Before:
API timeout
Malformed JSON
Dead source link
Prompt too large
After:
Retry 2 succeeded
JSON repaired
11/12 links verified
Context reduced 38%
Failure metadata attachedFor researchers, compliance teams, students, and engineers who need answers from private PDFs, docs, and source code without guessing.
Question:
Where does the app validate license keys?
Repo Intelligence:
lib/license-keys.ts
app/api/license/validate/route.ts
tests/license-keys.test.ts
Local Doc RAG:
Policy source: security-runbook.pdf
Relevant section: offline activation rulesEach project should feel like it belongs to its user and job: secure vault, coding workbench, marketplace, code map, document research desk, and practical business dashboard.
Premium privacy assistant: encrypted memory, local vault, and user-controlled workflows.
Practical coding agent: reads repos before writing code and shows the diff.
Agent skill economy: browse, price, install, and trade capabilities.
Architecture-first interface for understanding private source code with citations.
Calm knowledge UI for private PDFs, DOCX files, citations, and source excerpts.
Dense small-business pipeline UI for contacts, follow-ups, and local AI sales help.
Not every project needs to connect to every other project. The goal is to make the strongest work easy to understand, then use the right proof for the right audience.
HammerLock, Craig, and the infrastructure repos show product thinking, full-stack execution, privacy, reliability, and agent design.
The commercial wedge: $500 starter sites, lead capture, CRM, follow-up, reviews, and approval-first social drafts.
Daily build logs, short demos, anonymized business audits, and repo explainers turn the portfolio into a visible proof stream.
Private AI workspace for sensitive work: encrypted memory, local vault, specialist agents, and reusable context.
RecallMax, Tool Use Guardian, Prompt Condenser, Real-Time Verifier, Surgical Editor, and Code Reviewer solve agent failure points.
Genesis and other exploratory projects show curiosity, ambition, and comfort building outside one narrow lane.
Each repo answers a practical question: who would use this, what pain does it remove, and what proof can someone watch without running the code?
Lawyers, founders, finance teams, journalists, and operators use it when they need AI help on sensitive work without sending private context into a generic cloud workflow.
Solo founders, small teams, and agencies use Craig to inspect a repo, plan a feature, edit targeted files, run checks, and hand back a diff instead of vague coding advice.
Local businesses use it to turn offers, reviews, seasonal promos, and owner notes into approval-first posts for X, LinkedIn, Facebook, Instagram, and Google Business.
Agent builders use it when conversations, projects, or customer history outgrow the context window and the agent needs compact, retrievable long-term memory.
Production AI teams use it to wrap brittle tool calls with retries, timeout handling, schema checks, and repair paths so an agent fails less silently.
Startups and high-volume LLM apps use it to reduce prompt size, control token spend, and preserve important structures like URLs, code blocks, and task intent.
Newsrooms, research tools, and agent workflows use it to check URLs, validate generated claims, score confidence, and catch broken outputs before they move downstream.
New engineers, team leads, and coding agents use it to ask source-grounded questions about a codebase before changing anything expensive.
Researchers, compliance officers, students, and operators use it to query PDFs and internal documents locally with source-backed answers.
Coding agents use it when they need precise patches instead of rewriting entire files, reducing diff noise and lowering the risk of breaking unrelated code.
Solo developers, bootcamp students, and CI pipelines use it to catch security, quality, and maintainability issues before merge.
Competitive research tools, agents, and analysts use it to turn web pages into cleaner text, links, metadata, and source material for downstream workflows.
High-volume document workflows use it when they need fast, offline extractive summaries without paying an LLM for every preprocessing step.
Small businesses use it to track leads, follow-ups, open deals, customer notes, and local AI-assisted sales workflows without a heavy enterprise CRM.
Agent platforms use it as a speculative marketplace layer where agents can discover, price, install, and exchange reusable capabilities.
Websites, CRM, and AI follow-up for local businesses. Craig finds the digital gaps, builds the missing stack, and gives busy owners a simple system they can actually use.
Craig scans a business website or public listing and turns the gaps into a simple action plan.
Fast one-page or small-site build with services, trust signals, calls to action, and local SEO structure.
Contact forms, quote requests, appointment links, missed-lead tracking, and clean intake fields.
SynthPipe-style pipeline for new leads, follow-ups, booked calls, open estimates, and won customers.
Email templates, review requests, reactivation messages, and next-step suggestions for busy owners.
A modern services site with appointment CTAs, insurance/payment notes, review prompts, missed-call follow-up, and local SEO pages.
A fast mobile site with service area pages, quote forms, photo proof, job-type routing, and follow-up reminders for open estimates.
A polished offer page, treatment menu, before/after gallery, lead magnet, reactivation emails, and simple campaign tracking.
A trust-focused site with tax-season intake forms, service pages, document checklists, lead status, and automated next-step emails.
A professional site with practice-area pages, consultation forms, source tracking, intake triage, and careful privacy-aware messaging.
A compliance-aware site with service explainers, market-entry forms, gated downloads, CRM tracking, and multilingual content.
One-page website, contact form, local SEO basics, and a simple CRM pipeline.
Hosting, updates, lead tracking, review prompts, and follow-up templates.
Content, landing pages, automations, campaign ideas, and monthly improvement audits.
Narrated demos that show the work instead of asking people to believe the pitch.
These players include narration. Press play, then check the speaker icon inside the video controls and the browser tab volume. Links marked silent preview intentionally have no narration.
Drafts platform-specific posts for Craig, customer businesses, and operator outreach. Approval-first, no passwords, no auto-posting.
Shows the local-business stack people can actually buy: website score, lead capture, follow-ups, reviews, and monthly action.
One quick tour showing the body of work: local-first AI, agents, business automation, and deployed product surfaces.
More proof clips for the projects where a demo adds value. Smaller repos are grouped into system demos when that tells the truth better.
Private research, verification, memory, and vault output.
Reads the repo, plans, edits, checks, and reports the result.
Guardian, Verifier, and Condenser make agents less fragile.
Retries, timeouts, JSON repair, and failure metadata for agent tools.
Compress large prompts while preserving URLs, code, numbers, and intent.
Check links, structure, confidence, and outputs before downstream use.
Targeted patches and cleaner diffs for coding agents.
Source-grounded answers before code changes.
Ask questions across local PDFs and private documents.
Compress and retrieve useful long-term agent context.
Pipeline, follow-up, and local AI-assisted business workflows.
Scrape, summarize, compress, and prepare cleaner agent context.
Agents discovering, pricing, and installing reusable skills.
This is the shape of the work: local-first AI that can understand private context, use tools, verify outputs, preserve memory, and help build the next thing.