Role: Product Manager & Builder
PMs spend half a day on average doing competitive analysis, manually opening dozens of tabs with data scattered everywhere, and the results are hard to keep up-to-date.
Build an AI-powered real-time competitive analysis dashboard with natural language queries, so PMs can get insights simply by asking.
Reduced competitive analysis time from 4 hours to 30 minutes, with natural language queries and auto-generated comparison reports.
Type "Compare pricing strategies of A and B" to instantly generate structured analysis results — no query syntax needed.
Set up a watch list and the system automatically updates daily on feature changes, pricing adjustments, and market dynamics.
One-click multi-dimensional comparison tables covering features, pricing, user reviews, market positioning, and more.
Automatically pushes notifications when competitors have major feature updates, pricing changes, or funding news, ensuring you never miss critical intel.
Rapid prototyping was the top priority. Streamlit let me quickly build an interactive dashboard within the Python ecosystem without needing a separate frontend framework. More importantly, the PM team can modify and extend features themselves, reducing dependency on the engineering team.
Competitive analysis requires processing large volumes of long-form text (product pages, reviews, news). Gemini's long context window (100K+ tokens) can handle complete information in one pass, avoiding context loss from chunked processing. Additionally, it's more cost-effective than GPT-4, making it suitable for high-frequency automation scenarios.
We used a Role + Dimensions + Format framework to ensure output quality. First, define the AI's role (senior market analyst), then specify analysis dimensions (features, pricing, market positioning, user reviews), and finally standardize the output format (structured tables + key insight summaries), ensuring consistent and actionable results every time.
This project taught me that modern PMs don't need to wait for engineering resources. With no-code/low-code tools and AI, PMs can validate ideas, build prototypes, and even deliver usable products on their own. The key isn't technical prowess — it's being able to define the problem clearly and find the fastest path to validation.
The biggest challenge during development wasn't the AI model — it was data structuring and cleaning. Web-scraped data varies wildly in quality, directly impacting AI analysis accuracy. We ended up spending about 40% of our time optimizing the data pipeline, which is the most commonly underestimated component of many AI products.
The initial version only had basic query functionality, but after internal testing we discovered that what PMs needed most wasn't "querying" but "tracking" — they wanted to set up a competitor watch list and automatically receive daily change reports. This insight led us to reposition the product's core value proposition from "AI search tool" to "AI competitive monitoring platform."
Interested in this project? Let's connect.