The Lab

A working archive of ideas and tools.

Study notes, technology, visual work, and the references that feed the professional practice.

Technology in practice

Tools are useful when the system around them works.

A practical stack for research, reasoning, building, and shipping dependable work.

AI tools: 13
Capabilities: 68
Ecosystems: 5

The working stack

Selected for context, control, and how well each tool connects to the next stage of the work.

Connected workflow

Vercel V0

GitHub Copilot

Google Gemini

Anthropic

Claude Code

xAI Grok

OpenAI

VS Code IDE

NotebookLM

Dia

Perplexity

Google AI Studio

Hermes Agent

Capabilities, grouped by use

Depth matters more than a long logo wall. The emphasis below reflects where the stack does real work.

Languages & Frameworks

PythonPrimary
TypeScript / Next.jsWorking
SQLPrimary
RWorking

Finance & Legal Tools

Bloomberg TerminalPrimary
Excel / VBAPrimary
LexisNexisPrimary
FSCA Regulatory DocsPrimary

Infrastructure & Productivity

Git / GitHubPrimary
VercelWorking
Notion / ObsidianPrimary
SupabaseWorking

Frontend Frameworks

ReactPrimary
Tailwind CSSPrimary
Framer MotionWorking
Vue.jsWorking
SvelteExploring
Next.jsWorking

Backend Frameworks

Node.jsWorking
Express.jsWorking
Next.js API RoutesWorking
FastAPIWorking
tRPCWorking

Databases

SupabaseWorking
MongoDBWorking
RedisWorking
PrismaWorking
PostgreSQLPrimary

Testing & Quality Assurance

JestWorking
PlaywrightWorking
CypressWorking
ESLintPrimary
VitestWorking

Build Tools & DevOps

ViteWorking
GitHub ActionsWorking
DockerWorking
VercelWorking
TurborepoWorking

Embedding & Vector AI

Voyage AIWorking
pgvectorWorking
Nomic EmbedWorking
Cohere EmbedWorking
OpenAI EmbeddingsWorking

Agent Protocols & MCP

ValibotWorking
Agent-to-AgentWorking
MCP SDKsWorking
ZodPrimary
Model Context ProtocolPrimary

TanStack Ecosystem

TanStack VirtualWorking
TanStack RouterWorking
TanStack FormExploring
TanStack QueryWorking
TanStack TableWorking

AI Orchestration

LangChain.jsWorking
InngestWorking
MastraWorking
Vercel AI SDKPrimary
LlamaIndex TSExploring

UI & Component Libraries

Headless UIWorking
shadcn/uiPrimary
VaulWorking
Radix UIPrimary
cmdkWorking

Edge & Cloud Infra

Trigger.devWorking
Vercel Edge FunctionsWorking
Cloudflare WorkersWorking
Upstash RedisWorking
Supabase Edge FunctionsWorking

Systems, not silos.

Each ecosystem covers a complete path from source material to a usable outcome.

Prompt-to-Production Pipeline

From natural language prompt to deployed application in a single workflow. v0 generates UI, Copilot fills implementation gaps, Vercel handles deployment.

Idea
v0 Prototype
Copilot Refinement
Vercel Deploy

Research-to-Insight Engine

Legal and financial research starts with Perplexity surfacing the latest sources, NotebookLM grounding synthesis in uploaded documents, then flows through Bloomberg and LexisNexis for validation before producing actionable insight reports.

Perplexity Search
NotebookLM Synthesis
Bloomberg/Lexis
Insight Report

Data-to-Decision Framework

Raw data enters through SQL queries, undergoes statistical analysis in Python or R, and surfaces as interactive visualisations in Next.js dashboards.

Raw Data
SQL Queries
Python/R Analysis
Dashboard

AI Discovery Pipeline

From open question to cited brief in a single session. Perplexity surfaces the landscape with real-time citations, NotebookLM grounds the analysis in your uploaded corpus, and Dia provides ambient context as you browse, producing structured, source-verified briefs.

Question
Perplexity Search
NotebookLM Grounding
Structured Brief

Agent Skills & Harness Engineering

Reusable SKILL.md files encode when a capability activates, the exact steps to follow, and what to verify before handing off output. Wired into a harness like Claude Code or Spring AI, a single agent composes multiple skills per task, giving every run the consistency of a human SOP without manual orchestration.

SKILL.md
Harness Wiring
Skill Composition
Consistent Output

How the work stays sound

A compact operating method for using fast tools without giving up judgment, traceability, or quality.

Context-First Development

Every tool is chosen for its contextual fit, not its popularity. The right abstraction at the right layer reduces friction and compounds productivity over time.

Rather than forcing a single framework onto every problem, each project begins with an assessment: what are the constraints, who are the users, and what does the data look like? Python for analysis, TypeScript for interfaces, SQL for persistence. The tools serve the problem, never the reverse.

AI as Infrastructure

AI tools are not novelty features but foundational infrastructure integrated at every stage of the workflow, from research to deployment.

LLMs assist with legal research before a single document is drafted. Copilot accelerates implementation. v0 prototypes interfaces. NotebookLM grounds analysis in source documents without hallucination. Perplexity delivers real-time, cited answers for regulatory and market intelligence. Dia weaves ambient AI into everyday browsing. The entire development lifecycle is augmented, producing higher quality output with tighter feedback loops.

Data-Driven Decisions

Professional expertise is amplified by data, not replaced by it. Quantitative tools provide the evidence base for qualitative judgment calls.

Bloomberg data informs investment thesis. SQL queries surface patterns in regulatory filings. Statistical analysis in R validates hypotheses before they reach stakeholders. Every recommendation is anchored in evidence.

Security & Compliance by Default

In finance and law, security is not optional. Every tool selection considers data handling, access controls, and regulatory compliance from day one.

Supabase for row-level security. GitHub for audit trails. Self-hosted models where client data sensitivity demands it. The stack is built to satisfy the scrutiny of financial regulators and legal ethics boards.