Live tools, documented experiments, and data pipelines — each with a GitHub repo and real methodology.
AEO Source Classifier + Visibility Tracker
A live Streamlit tool that takes a source inventory CSV from any AI-search tracking experiment and surfaces source types, content shapes, domain persistence, and citation patterns — then generates a prioritised content action plan. Built from a real multi-day AEO experiment, not a hypothetical demo.
Key insight surfaced by the tool
URL volatility does not mean source-pattern volatility. The exact URLs changed across snapshots, but the source types and content shapes were consistent. The stronger strategy is building the kind of content the answer engine repeatedly trusts — not chasing a single citation slot.
4snapshots analysed
Livedeployed on Streamlit
Pythonpandas · Plotly
StreamlitPythonpandasPlotlyAEOSource classificationCitation trackingContent gap analysis
AEO Source Ecosystem Workflow — 7-Day Experiment
The research experiment behind the tool above. A structured 7-day AEO experiment: four Perplexity snapshots on a target buyer-intent query, a parallel distribution experiment across five organic channels, and a Claude Code-powered workflow that classified 40 citation instances, extracted source-shape patterns, mapped 8 adjacent query clusters, and produced a 30-day content architecture plan.
Non-obvious mechanic found
Answer engines reward source shape over domain authority. A smaller competitor held a citation slot across all 4 snapshots using a single, query-aligned product page. Platform trust also dominates semantic relevance — a relevant comment still fails if the platform doesn't keep it visible.
40citations classified
4Perplexity snapshots
8query clusters mapped
5channels tested
Perplexity AEOSource-pattern analysisClaude CodeBuyer intent mappingLinkedIn PulseQuoraSubstack
Marketing Attribution Simulator
A Streamlit app that runs four attribution models — first-touch, last-touch, linear, and time-decay — side by side on the same customer journey dataset. Shows how channel revenue rankings and ROI shift depending on which model is applied, with auto-generated insights and a downloadable action plan.
Key insight
Last-touch may credit Paid Search with 60% of revenue — but Linear and First-touch often reveal that Email drove the first engagement for every customer who eventually converted. Cutting Email would quietly drain the pipeline that Paid Search closes. Attribution models are lenses, not ground truth.
4attribution models
1,057touchpoints in sample
Livedeployed on Streamlit
StreamlitPythonpandasNumPyPlotlyCampaign attributionChannel ROIFunnel analysisMarketing analytics
Customer Sentiment & Marketing Performance Analytics
Full end-to-end data pipeline: raw CSV data into SQL Server (Docker), Python sentiment analysis using NLTK VADER, and an interactive Power BI dashboard published to Power BI Service. Built to surface why customer satisfaction scores were declining despite increased marketing spend.
Key finding
Review text sentiment doesn't align with star ratings — qualitative feedback adds critical context beyond numeric scores. Social media showed high engagement volume but weak conversion, suggesting a channel attribution problem before scaling spend.
3tool layers
5dashboard pages
VADERNLP model
SQL ServerPythonNLTK VADERPower BISentiment analysispandasDockerFunnel analysisData validationData governanceStakeholder reporting
Inbound Lead Qualification Agent
An AI-powered qualification workflow that turns raw inbound leads into a CRM-ready output — with lead scores, fit levels, qualification reasons, recommended next actions, and draft follow-up emails. Also includes n8n automation assets for extending the workflow into a fully automated pipeline with webhook triggers, CRM updates, and Slack notifications.
Business impact
Reduced estimated lead review time from 50 minutes to 5 minutes across a sample workflow — 10 min per lead manually vs 1 min automated. Identified 3 high-fit leads out of 5 with qualification reasoning attached to each.
10xfaster lead review
n8nautomation ready
Livedeployed on Streamlit
PythonStreamlitpandasn8nLead scoringCRM simulationRevOpsMarketing opsWorkflow automation
