My Projects
A deep dive into the problems I identified, the users I served, and the product decisions behind each project, including the architecture, GTM strategy, and outcomes.
ChatMonkey
PM Framing
Problem: Small businesses can't afford 24/7 phone support and lose leads outside business hours. Current platforms didn't help to understand what customers were asking about and how to improve.
Vision: Enable businesses to effortlessly capture leads and provide instant, 24/7 customer support through custom-trained AI chatbots.
PM Actions: Defined product vision and roadmap; conducted user interviews with small business owners to identify key onboarding friction; designed the subscription model and pricing tiers; managed the full build lifecycle from architecture to Stripe integration.
Outcome: Live SaaS product at mychatmonkey.com with active users; currently user testing and iterating.
Architecture & Execution
Technology Utilized
Implementation Details
- Built a full-stack SaaS platform allowing users to create, train, and embed AI chatbots in minutes.
- Developed a responsive dashboard using React and TailwindCSS for bot management, conversation history, and analytics.
- Engineered a scalable serverless backend on AWS to handle real-time chat processing.
- Integrated Stripe for seamless subscription billing and plan management.
- Utilized the OpenAI API to power intelligent, context-aware responses.
Gallery
Summit County Housing Analytics
PM Framing
Problem: Mountain real estate buyers and sellers lack reliable predictive tools. existing market data is lagging, opaque, and not localized to micro-markets.
Vision: A production-grade analytics platform that combines Advanced SQL Engineering with Deep Learning to analyze and predict real estate trends in my home area of Summit County, CO.
PM Actions: Defined the product vision and data requirements; architected the full pipeline from raw MLS data to ML model; prioritized explainability (SHAP) as a core product feature to build user trust; designed the Streamlit dashboard as a validated interactive demo.
Outcome: Production-grade ML platform with an interactive live demo; model achieves high-accuracy price predictions on Summit County data.
Architecture & Execution
Technology Utilized
Implementation Details
- Built a normalized SQLite warehouse using complex window functions and CTEs for trend analysis.
- Developed a custom PyTorch feed-forward architecture for high-accuracy property price prediction.
- Implemented Explainable AI (XAI) using SHAP values and Partial Dependence Plots to interpret model drivers.
- Engineered a resilient data pipeline with a Dead Letter Queue (DLQ) pattern and strict Pydantic validation.
- Created an automated "Champion/Challenger" model registry with temporal cross-validation.
- Deployed an interactive Streamlit dashboard featuring uncertainty estimation via Quantile Regression.
Gallery
TextMeAvy
PM Framing
Problem: Overnight winter backcountry travelers lose cell service and have no way to check current avalanche forecasts. Several 2026 incidents have highlighted this gap.
Vision: Make critical avalanche safety data accessible to backcountry travelers regardless of cell coverage, via the satellite messengers they already carry.
PM Actions: Identified the market gap by extending TextMeWeather to a safety use case; conducted user interviews with backcountry travelers; defined the two-way satellite SMS pipeline as the core architectural requirement; designed the onboarding journey based on beta user feedback; segmented users (ski tourers, snowmobilers) for targeted marketing.
Outcome: Live in Apple App Store and Google Play; 50% month-over-month user growth; zero-budget GTM via targeted community outreach.
Architecture & Execution
Technology Utilized
Implementation Details
- Built React frontend to emulate the widgets available from US Avalanche Centers such as the CAIC.
- Developed apps for iOS and Android, hosted in Apple App Store and Google Play Store.
- Developed serverless AWS backend to host APIs, authentication, and website.
- Built vanilla HTML/JS websites hosted on AWS S3 with authentication via AWS Cognito and subscription management via Stripe.
- Built a system to receive text-based requests and automatically generate and send visual forecast outputs (like elevation maps and pinwheels) back to the user.
- Developed a complete two-way communication pipeline to bridge satellite messaging protocols with web-based avalanche data.
Gallery
TextMeWeather
PM Framing
Problem: Satellite messengers like Garmin inReach provide a single weather icon for the whole day. Backcountry travelers need detailed, location-specific forecasts before they lose cell service, and there was no good way to get them.
Vision: Let any backcountry traveler request detailed weather forecasts via satellite messenger and get a useful, formatted response back, with no cell signal required.
PM Actions: Prototyped the core idea; designed the request/response format based on satellite messenger constraints; tested the prototype on the trail; interviewed beta users (backpacking guides, outfitters) to refine UX; led targeted marketing outreach with zero paid budget; iterated roadmap from user feedback on onboarding friction.
Outcome: 10+ MAU within first month, zero marketing budget; live in Apple App Store and Google Play; 50% month-over-month growth.
Architecture & Execution
Technology Utilized
Implementation Details
- Launched in June 2025 to beta users with outbound marketing to backpacking guides and power users.
- Achieved 10+ MAU user base within a month using no marketing budget.
- Developed apps for iOS and Android, hosted in Apple App Store and Google Play Store.
- Developed serverless AWS backend to host APIs, authentication, and website.
- Built vanilla HTML/JS websites hosted on AWS S3 with authentication via AWS Cognito and subscription management via Stripe.
- Interviewed users to refine website and app UX.
- Created automated deployments using CloudFormation.
- Deployed to two regions with a hot-backup strategy.
- Architected an event-driven SMS pipeline (AWS Lambda + SQS) to parse unstructured satellite messages, ensuring reliable two-way communication.
Gallery
Snotel.Info
PM Framing
Problem: The official SNOTEL data interface is technical and hard to use, so skiers and outdoor enthusiasts couldn't quickly access the snowpack information they needed.
Vision: Provide a modern, user-friendly dashboard for SNOTEL snow and weather data to make critical snowpack information instantly accessible.
PM Actions: Defined the product as a free, frictionless alternative to the USDA interface; designed the map-based UX to surface key metrics at a glance; used it as a beachhead GTM tool to attract attention to the Snow Intel paid product; refactored in 2025 to a modern serverless AWS stack as an open-source contribution.
Outcome: Active and maintained at snotel.info; served as a freemium funnel to attract Snow Intel beta users; open-sourced on GitHub.
Architecture & Execution
Technology Utilized
Implementation Details
- Built simple node.js server and HTML website as a dashboard for SNOTEL data from the USGS.
- Developed Progressive Web Application (PWA) functionality to allow site to function as an app on device.
- Utilized cron job and processing script to create time-aggregated metrics, available within a map layer.
- Created and distributed for free to attract attention to Snow Intel.
- Refactored in 2025 to serverless AWS hosted API and website. Repo is available open-source: GitHub.
- Leveraged AWS CloudFormation to quickly deploy S3, Lambda, API Gateway, and EventBridge resources.
Gallery
Snow Intel
PM Framing
Problem: Backcountry skiers had no way to understand what was happening inside the snowpack before heading into the field, a core safety gap that existing weather apps couldn't solve.
Vision: Help backcountry skiers understand snow conditions for route planning with snowpack structure, weather forecasts, and satellite imagery.
PM Actions: Researched the market and existing substitutes; architected the full backend on self-hosted hardware; hired and managed a 3-person contractor team on Upwork to deliver the React frontend in 3 months under $2,500; created a launch plan with digital ads and in-person promotions; ran user interviews and moderated usability tests; identified root cause friction and iterated the wind model; made the disciplined decision to retire the product when a physics constraint (not a software one) prevented further improvement.
Outcome: Attracted 30+ beta users; 100% lift in session duration after feature iteration; deliberately retired after recognizing an insoluble physics bottleneck, one of the most valuable PM lessons of my career.
Architecture & Execution
Technology Utilized
Implementation Details
- Utilized open-source library Alpine3D and NOAA HRRR model outputs to predict layers in the snowpack.
- Obtained a static IP via Comcast and domain via Squarespace. Built Squarespace website backed by Google Analytics.
- Built and self-hosted processing, database, API, and tile servers.
- Architected weather data processing workflow to ingest weather data, produce snowpack predictions, and store <500GB of forecasted data per day.
- Managed a team of three contractors to produce a React frontend within 3 months for <$2500 budget.
- Created a launch plan to target beachhead users with digital ads, in-person promotions. Attracted >30 beta users. Retained several users for >3 months.
- Gathered user feedback through user interviews and moderated usability tests.
- Determined that wind deposition of snow was causing huge variance in user results. Created a refined wind model in OpenFOAM to improve results.
- Retired the product after identifying that micro-climate wind modeling was a physics constraint, not a software one. A key lesson in recognizing insurmountable technical hurdles.