TLDR: Conceived and prototyped an NER based free text search system (Project Morpheus) overnight as an alternative to a failing AI initiative. After months of advocacy and setbacks, I built the full product from scratch in one week after returning from paternity leave and delivered Realtor.com’s most impactful launch of the year. Morpheus increased lead conversion 7× over legacy search, established a sustainable foundation for future AI products, and added negligible operational cost.

1. Context & Stakes

Over the past few years, Realtor.com had been exploring ways to leverage AI to create a differentiated search experience. One of the flagship efforts was an "AI Search" initiative intended to help users find exactly what they wanted through more intelligent, contextual queries (search at the time consisted of searching for a geographic area, then using dropdown filters to narrow down results).

Despite nearly two years of investment, the project struggled to meet quality bars and failed to ship a viable product. Marketing campaigns were paused, leadership confidence was fading, and competitors, particularly Zillow and Homes.com, were beginning to gain momentum with their own AI features.

By Fall 2024, the initiative was on the verge of collapse. User tests returned CSAT scores below 3/10, and engineering velocity had stalled.

2. Problem → Insight

The project’s direction had drifted toward complexity. The team was investing heavily in custom ML pipelines and an LLM designed to scan listing photos for features a user might describe. For example, “a house with a white picket fence.”

After stepping back to evaluate the problem (my team owned the front end experience), I realized both the product strategy and technical approach were misaligned with user expectations and system realities.

• Real usage data told a simpler story. Analysis of filter interactions showed that our existing structured filters already covered ~98% of user intent. Generating image based metadata on demand was over engineering for negligible gain.
• Performance and cost were unsustainable. LLM inference added multi second latency and high per-query cost, unacceptable for a product where users expect sub second search results, and we process tens of millions of requests a day.
• The UX was fractured. The current AI prototype took users out of the normal search flow, introducing friction and reducing engagement.

The insight was clear:

The idea of "just typing what you want" was right, but the approach was wrong.

We didn't need a new AI heavy search engine; we needed a lightweight, natural language layer that mapped free form text to our existing, high performance filter system.

Realtor.com handles tens of millions of searches per day; the winning solution had to feel instant, scale efficiently, and operate at near zero incremental cost.

3. The Overnight Prototype

That same night, I built the first working concept of what I believed was the right path forward, a lightweight JavaScript parser paired with a React demo front end.

The parser used a Named Entity Recognition (NER) approach combining regex and JavaScript logic to extract intent from user queries. A small subset of our filters required special handling (min/max ranges, numeric bounds, etc), but the vast majority of our ~300 filters were boolean. This made it feasible to capture common phrases and synonyms and translate them directly into our filter schema.

It could parse queries like "3 bedroom with hardwood floors under 600k" and map them to structured search filters in milliseconds.

The next morning, I demoed the prototype to our PM. He immediately saw the potential, and together we began socializing the concept. Over the next several weeks, we secured leadership buy in through live demos, including one presented to the executive team showing how our approach outperformed competitor offerings. Realtor.com has been reactionary to Zillow for years, but this was our chance to beat them to market with a faster, smarter, more scalable solution.

This rapid proof of value shifted internal sentiment overnight, from a demoralized "failed AI project" to an achievable, high impact product.

4. Approach & Execution

Just as the project was greenlit, I went on paternity leave. When I returned, the lead developer had left the company and the codebase was nonfunctional. After a rapid evaluation, it became clear that the approach taken was fundamentally flawed and would need to be rebuilt from the ground up.

With only four weeks remaining before a hard marketing deadline, I took ownership of the project and rebuilt the entire codebase from scratch in one week. The new version met the quality bar product leadership had set, and I went on to lead the engineering team in polishing the feature and preparing it for launch.

To accelerate delivery, I divided execution into four parallel workstreams:

• Testing: Built a robust suite of test cases to validate how accurately the parser matched real world user inputs.
• Filters: My initial rebuild included the core logic but only a subset of filters. The team expanded coverage to all ~300 filters, including synonym handling and range logic.
• Bugs: Organized company wide bug bashes to uncover edge cases and improve stability.
• Validation: Integrated A/B testing, telemetry, and analytics tracking to measure quality and performance.

To maintain momentum, we used lightweight code reviews and daily cross discipline syncs. I continued to contribute heavily to both front end and parser logic while driving technical decisions, rollout strategy, and overall production readiness.

5. Design & Technical Highlights

The final implementation of Morpheus balanced precision, speed, and maintainability. Every design choice was optimized for scalability and user experience at minimal operational cost.

• Natural-Language Parsing: Developed a custom tokenization and heuristic mapping layer that extracted location, price, and attribute context with over 97% accuracy in test queries.
• BFF API Layer: Exposed the parser through a lightweight backend for frontend (BFF) API, enabling use across both web and native clients with a single logic layer.
• Performance: Achieved ~40 ms latency per request with no additional backend resources required, adding negligible operational cost to handle existing query volume.
• User Interface: Introduced a redesigned search bar that encouraged free form input and provided real time parsing feedback to guide users as they typed, as well as communicating what terms we did not understand.
• Accessibility: Ensured full keyboard navigation and WCAG 2.1 AA compliance, matching Realtor.com’s accessibility standards.
• Observability: Integrated structured logs and performance tracing to monitor latency, accuracy, and user interaction depth.

6. Impact

Within days of launch, the results were undeniable.

Morpheus vs Legacy Search

Morpheus vs Original AI Approach

Morpheus became Realtor.com’s largest release of the year, featured in marketing campaigns and internal all hands as a model of product velocity and engineering leadership.

7. Leadership & Cross-Functional Wins

• Built executive confidence by delivering a tangible prototype within 24 hours and reframing a stalled initiative into a viable product.
• Unified Product, Design, and Engineering around a single, achievable goal and maintained alignment through rapid iteration cycles.
• Provided daily updates to the SVP of Product and VP of Engineering, consistently connecting technical decisions to measurable business impact.
• Established a repeatable delivery model now used across multiple org initiatives: rapid prototype → measure → ship.

8. Risks & Mitigations

9. Aftermath & Lessons

Morpheus shipped ahead of competitors and restored confidence in our ability to move fast without breaking discipline.

The project demonstrated that clear product focus, lean engineering, and tight feedback loops can deliver innovation at enterprise scale, without the overhead of traditional "big AI" efforts.

What I’d Do Next

• Expand functionality and scalability: Now that we’ve beaten competitors to market, focus on deeper NLP integration to strengthen areas where rule based parsing has limitations.
• Leverage real user data: Continuously analyze production queries to identify gaps and inform backend AI tagging strategies.
• Add personalization: Introduce contextual memory for repeat users to make search feel adaptive and self learning.
• Increase adoption through education: Improve in product messaging to teach users what's possible with free text search. With such high LSR, expanding adoption directly drives revenue.
• Extend as a platform service: Expose the parser as a reusable API powering additional consumer surfaces across the Realtor.com ecosystem.

10. Credits

Product: Matt Holihan
Engineering: Victor Cho, Ben Spears, Martin Robledo, Kuber Singh, Michael Pratt, and our leader Francisco Ovalle-Martínez
TPM: Anna Rai
Marketing: Meghan Ruff

Special thanks to everyone who contributed to making Morpheus a reality, your collaboration and energy made Morpheus possible.

Speed is strategy. Execution is innovation.