AI Tools
01
QueryGPT MVP
Python · FastAPI · SQLite · Anthropic API
Lets you ask a database a plain-English question and get the right answer back — no SQL knowledge needed. Inspired by how Uber built a similar internal tool.
- Caught 4 real bugs through live, unscripted testing
- Includes a simple web interface with a demo mode
02
Spec Audit Tool
Node.js · Vanilla JS · Anthropic API
Reviews a product spec before it gets built and flags what's missing — edge cases nobody thought of, requirements that can't be tested, numbers with no source behind them.
- Runs 18 separate checks against a spec
- Refined across 7 rounds of real testing, fixing 8 bugs along the way
03
MeetingMeet
Python · http.server · Anthropic API
Takes a meeting transcript, figures out the action items, decides what's safe to share vs. what should stay private, and drafts emails using only the safe parts.
- Tested against 12 real meeting transcripts with hand-checked answers
- Includes a working web app with export to CSV and PDF
04
Instagram → YouTube Song Sync
Python · SQLite · YouTube API
Takes music saved on Instagram and automatically builds a matching YouTube playlist — no manual searching.
View on GitHub →05
Scrum RAG Chatbot
Python · FastAPI · Anthropic API
A chatbot that answers Scrum questions by actually reading the official Scrum.org guides first, rather than guessing.
- Grounded in 4 official Scrum.org guides
- First project built end-to-end without a developer background
06
AI Agent Incident Audit
Python · FastAPI · Deterministic checks (no model in detection)
Checks whether an AI agent actually stayed within what it was told to do, using hard evidence from its actions instead of asking the agent to explain itself. Tested against 5 scenarios modeled on real 2026 agent incidents.
- Never trusts the agent's own explanation — only looks at the recorded trace of what it actually did
- Includes 2 control cases to prove it stays quiet on legitimate runs, not just catches violations
PMO / Business Tools
07
VendorPulse
React · Claude API · Risk Scoring · Governance · AI
Scores how risky each vendor is across six dimensions, tracks when governance reviews are due, and drafts committee-ready briefing documents at the click of a button.
Read case study → · View on GitHub →08
PropelIQ
React · Claude API · Benchmarking · Commercial Proposals
Generates hypothesis-driven consulting proposal outlines grounded in industry benchmark data — two connected modules that turn a data gap into a defensible value hypothesis, as one end-to-end workflow.
- Benchmark data feeds directly into the proposal's value hypothesis via a connected handoff
- Includes a Storytelling Check that catches fabricated specificity, not just internal flow
09
BriefCast
React · Claude API · PMO Reporting · Stakeholder Communications
Generates governance-ready status reports from raw JIRA sprint exports, escalation logs, and delivery metrics — with an interpretive layer that surfaces patterns the data implies, not just what it states.
- Five guardrails built in: data completeness, scope creep detection, RAG Red confirmation, tone consistency check, and audit trail
- Confidence flags (⚠️) surface ambiguous interpretations — human-in-the-loop by design
10
StoryStack
React · Claude API · Backlog Prioritization · TPM Tooling
Scores feature requests across Business Impact, Effort, and Strategic Fit — then assigns a Priority Band and one-line rationale per feature. Context-sensitive: the same backlog produces a different ranking under Growth vs Risk Reduction vs Compliance vs Efficiency.
- Four live demo runs across Payments and Data Platform domains confirm context-sensitivity
- Generates a 1-page Decision Brief formatted for engineering and leadership sign-off
AI Product Management Tools
11
FraudLens
Vanilla JS · Canvas API · Confusion Matrix · Financial Services AI
Interactive credit card fraud detection playground — adjust a detection threshold and watch the confusion matrix, cost curve, and business impact update in real time. Built to make AI model evaluation accessible to both technical and non-technical audiences.
- Live cost calculator showing fraud loss vs customer friction tradeoff
- Stakeholder persona toggle — same model through a Risk Officer vs CX Head lens
- Guided 4-step tour, traffic light indicators, and plain-language summary bar
12
ProbeLens
HTML · Claude API · First-Principles Framework · PM Tooling
Walks any product problem through an 8-row first-principles framework — generating sharp clarifying questions, flagging information gaps, and producing a locked one-page decomposition brief. Two modes: Brief Generator for structured analysis and Interview Practice with row-by-row scoring.
- 18 PM interview questions across 4 categories with batch scoring against the framework
- Style calibrated on 7 real PM problem samples — named entities, binary framing, plain English throughout
13
ValidIQ
React · Claude API · Discovery Scoring · Finserv PM
Scores how confident you should be in a discovery finding before it becomes a roadmap decision - flags weak assumptions, missing data, and ROI claims that haven't been tested.
- Runs 7 diagnostic questions across Clarity, Scale, Freshness, Accuracy, Data, and Constraints
- Generates a Claude-powered scorecard with a priority quadrant and next steps per finding
14
AI System Lens
HTML · Claude API · AI Design Matrix · Product Audit
Takes your AI product idea and tells you what questions to ask before writing a single line of code - scores it across 12 dimensions and locates it on a design matrix.
- Maps your product across Classical ML, LLM Design, and Agentic Design using a 12-question audit
- Audit scorecard flags what stage you are at and what stage to build for next
Scrum / Agile Tools
15
RetroLoop
Python · Claude API · Scrum
Spots when a complaint raised in a sprint retro is actually the same unresolved issue from a previous retro, just worded differently — so teams stop rediscovering problems they never actually fixed.
- Teams stop wasting retro time on issues that were already raised before but never properly closed
- Makes retro patterns visible across sprints, not just within the one you're currently running
16
Kanban Board Build
HTML · Claude API · Concurrency Testing
An investigation into what actually happens when two AI agents try to update the same Kanban board at the same time — one responding to chat, one running on a schedule. Real collision tests, documented findings.
- Helps teams understand the real risk of letting two AI agents write to the same data source without coordination
- Built as a reference case — so you know where the guardrails actually are before deploying multi-agent workflows