How Claude
Made Me a
10x Researcher
Hi, I'm
Nimer Singh.
I'm a Scam researcher - I spend my days hunting for the threats that try to reach everyday users on the internet.
Before this, I worked as a Lead Developer in an IoT and Industry 4.0 startup. I have enjoyed working with security for the past 6 years.
I find the
bad stuff before
you click it.
My team protects people from malicious websites — phishing pages, scams, fake stores, the lot. I research the patterns attackers use, then turn that research into signals and detections that block them at scale.
I turn haystacks
into answers —
as fast as questions arrive.
Investigate
Pull threads on suspicious sites and clusters of attacker activity.
Build
Prototype tools, scrapers, and pipelines fast.
Query
Interrogate big datasets and surface the anomalies.
Fix Bugs
Reproduce, isolate, and patch — without losing momentum.
Research is slow
by default.
Context Switching
Every investigation spans different languages, several dashboards, and a stack of half-remembered docs.
Boilerplate Tax
Most of the day is glued coding and rewriting the same query for a new data source.
Scale Problem
Manual review works for ten URLs. It collapses at ten thousand. The signal is always one zero away.
Enter
Claude.
Not as a chatbot. As a research partner that codes, queries, and reasons — from a single command line.
Right tool, right
scale, right now.
Small tasks.
Live tools.
- Quick lookups, triage, exploration
- Live tool calls in the chat loop
- Human-in-the-loop, low latency
- Best for tens to hundreds of items
Big jobs.
Claude scripts.
- Generate the analysis code, then run it
- Batch over thousands of records
- Reproducible, version-controlled
- Best for one-off scripts and bulk work
Four ways I get
leverage.
Investigate
Open a research thread. Claude pulls context from our data sources, cross-references signals, and points at the interesting edges — before I even open a notebook.
Build
From "I need a tool that does X" to a working prototype in minutes. For bigger jobs, multiple agents work in parallel while I review at the seams.
Query at Scale
Plain-English to SQL across our warehouses and search clusters. When the result set is huge, Claude writes a script to process it instead of trying to do it all in chat.
Debug
Share a stack trace and the repo. Claude reproduces, hypothesizes, and patches — a half-day rabbit hole becomes a 20-minute conversation.
One command line
to rule them all.
OpenSearch
Tooling
Teach it once.
Use it forever.
Skills are reusable instruction packs that Claude loads on demand — a folder with a recipe that encodes a workflow's best practices, gotchas, and house style.
Instead of re-explaining a process every time, I write it down once. Claude reads the skill when the work matches — and follows the recipe.
Threat Analysis Playbook
Encodes how I evaluate a new lead - what to check first, which dashboards to pull, and the questions that separate noise from a real campaign.
Query Translator
Plain-English to SQL across our warehouses - but with our column quirks, naming conventions, and partition gotchas baked in, so Claude doesn't get them wrong.
Regional Coverage Analysis
Given a geographic region, this skill pulls a representative URL sample from our data warehouse, scores each one through our reputation system, waits for enrichment to complete, re-checks for changes, and writes a structured gap-analysis report — all from a single prompt.
From signal to
shipped report — in one afternoon.
Signal
A cluster of look-alike domains starts surfacing in our telemetry. Pattern unclear. Volume: ~12,000 URLs.
Triage
I describe the cluster to Claude. It queries our search layer, returns the top features, and proposes a hypothesis.
Scale
Claude writes a script to fetch and feature-extract all 12k URLs in parallel — work I'd usually push to next week.
Ship
Findings written up, charts generated, dashboard updated. Detection logic drafted, reviewed, deployed.
Five rules that
compound.
Match the tool
to the scale.
Live tools for exploration. Scripts for scale. Don't ask a chat to chew through a million rows.
Write down
your workflows.
Anything you've explained twice is a skill. Capture it once and reclaim the next ten hours.
Run agents
in parallel.
For bigger jobs, multiple sub-agents handle independent chunks while you stay on the critical path.
Trust, but
verify.
Claude drafts; you review. The speedup is real, but responsibility for correctness still belongs to the researcher.
Stay in the
terminal.
One interface, one history, one place to debug. Your future self will thank you.
The job got bigger.
So did I.
Claude didn't replace the research. It removed the friction between curiosity and answer — and that's the whole game.