Why We Built Our Own Flight Search Engine (And What We Learned)

Two years ago, we set out to build a better flight search engine. Not because the world needed another metasearch site—there are plenty—but because none of the existing tools answered the question travelers actually ask: “What is the cheapest way to get from here to there, considering all my options?”

Google Flights only shows cash. Award search tools only show miles. Transfer partner calculators only show ratios. No single tool brought it all together. So we built one.

Here is what we learned along the way.

Lesson 1: Airline Data Is a Mess

Airlines distribute fare data through a patchwork of systems built over decades. GDS (Global Distribution Systems) date from the 1960s. ATPCO fare filings use a format designed before the internet existed. NDC (New Distribution Capability) was supposed to modernize everything but instead created a third, incompatible layer alongside the existing two.

Our first six months were spent building adapters for these different data formats—normalizing fare classes, tax breakdowns, baggage policies, and routing rules into a single, consistent schema that our frontend could render.

Lesson 2: Speed Requires Parallelism

A typical Pointify search fans out to 5–8 data sources simultaneously. Duffel NDC takes 500ms–2s. Google Flights scraping takes 2–4s. Award database queries take 1–3s. Transfer partner calculations take 100–500ms.

If we called these sequentially, a single search would take 10–15 seconds. Instead, we built the entire pipeline on async I/O with Python’s asyncio, launching all queries in parallel and streaming results to the browser via Server-Sent Events as each source responds.

The first results appear in under one second. The full result set is typically complete in 3–5 seconds. This required rethinking the entire UI to handle progressive loading—results arriving, sorting, and re-sorting in real time as new data streams in.

Lesson 3: Points Optimization Is a Graph Problem

When a user with Chase and Amex points searches for a flight to Tokyo, we need to evaluate every possible transfer path:

• Chase UR → United → direct booking
• Chase UR → Virgin Atlantic → ANA first class
• Chase UR → Singapore KrisFlyer → Star Alliance partner
• Amex MR → ANA → round-trip award
• Amex MR → Virgin Atlantic → ANA one-way
• …and dozens more combinations

Each path has a different transfer ratio, different award pricing, different availability, and potentially a transfer bonus that changes the math. We model this as a weighted directed graph and find the shortest paths (lowest cost) using modified Dijkstra’s algorithm. The graph is recomputed on every search to account for real-time transfer bonus changes.

Lesson 4: The Hardest Part Is Not Technical

Building the search engine was hard. But the hardest part was getting access to data. Airlines guard their pricing data carefully. NDC requires bilateral agreements with each airline. Award availability data is scattered across proprietary systems with no public API.

We solved this through a combination of legitimate API partnerships (Duffel, Seats.aero), browser automation for publicly available data (Google Flights), and direct relationships with airline loyalty programs. Every data source required weeks of negotiation, integration, and testing.

Lesson 5: Users Trust Transparency

Early user testing revealed something surprising: people did not trust our results. Not because the prices were wrong, but because they seemed too good. “Why is this business class flight only showing as $400 in points value when Google says the cash price is $3,000?”

We added the cents-per-point calculation, the transfer path visualization (“Chase UR → Virgin Atlantic → ANA”), and the comparison view (cash vs. points side by side) specifically to address this trust gap. Showing the math behind the recommendation increased user confidence by over 60% in A/B tests.

What We Are Building Next

The core search engine is live, but we are just getting started. On the roadmap:

• Fare prediction: ML model trained on historical pricing data to recommend whether to book now or wait.
• Hotel search: The same cash-vs-points comparison for hotel loyalty programs (Hyatt, Marriott, Hilton, IHG).
• AI booking agent: An agent that can actually complete the booking across airline websites, handling the complex multi-step processes that make award booking tedious.
• Price alerts: Real-time monitoring of both cash fares and award availability for your saved routes.

We are building in public and shipping fast. Follow the blog for technical deep dives, product updates, and the occasional mistake fare alert.