Speed Demon | 72-Hour Hackathon

About the Hackathon Speed Demon is a 72-hour online performance championship built around one challenge: make your code faster, leaner, and more efficient. Most hackathons ask a panel whether they liked your idea. Speed Demon measures whether your code actually performs. Every track provides a task specification, a dataset, and a deliberately naive reference implementation that solves the problem correctly and slowly. Your challenge is to solve the same problem correctly — and not slowly. The competition covers five tracks: log processing, search indexing, complex CSV parsing and joining, graph algorithms, and lossless compression. Each track tests a different aspect of performance engineering, from memory layout and parsing efficiency to shortest-path algorithms and compression trade-offs. This is not a hackathon where you submit benchmark numbers from your own laptop. We run your code ourselves. Every submission is built from your Dockerfile and executed on the same dedicated machine in Helsinki, with fixed CPU resources, memory limits, and a controlled benchmarking environment. Your output must match the reference implementation exactly, byte for byte, or you receive no score for that track. The competition uses a published efficiency coefficient that combines execution time, peak memory, source size, dependency weight, and scaling behaviour. Machine-measured performance accounts for 65% of each track's score, while human judges evaluate code quality, optimization technique, and your performance log for the remaining 35%. Participants can enter multiple tracks, compete for individual track prizes, and qualify for the overall Speed Demon title by completing at least three tracks. Every millisecond is a soul. Stay Updated All event announcements, track specifications, rule clarifications, dataset releases, scoring harness updates, leaderboard information, submission instructions, FAQs, and participant support are provided through the official Hackathon Raptors community channels. Why Participate Most performance competitions focus on raw execution time. Speed Demon takes a broader approach: the challenge is to produce software that is measurably faster, uses resources efficiently, scales predictably, and remains technically sound. Participants will get to: Optimize algorithms and data structures against fixed, measurable workloads. Process 3.1 million structured log lines and calculate latency percentiles and error rates. Build an inverted search index and answer 50,000 Boolean and phrase queries. Parse hostile CSV files containing quoted fields, escaped quotes, embedded newlines, and mixed line endings. Optimize shortest-path queries on a directed graph containing 5 million edges. Implement lossless compression for a 100 MB mixed corpus. Explore CPU performance, cache behaviour, memory allocation, data layout, and parallel execution. Practice profiling, benchmarking, and systematic performance optimization. Balance execution speed, memory consumption, source size, dependency weight, and scalability. Build reproducible, offline-capable Docker submissions. Compete on a public leaderboard using a transparent scoring formula. Document optimization decisions, including failed experiments and measured improvements. Demonstrate engineering skill through reproducible results rather than pitches or presentations. AI coding assistants are allowed and expected. The challenge is not whether AI helped. It is whether your implementation produces correct output, performs well under controlled conditions, and demonstrates sound engineering decisions. Eligibility Speed Demon is open to: Students. Software developers. Systems engineers. Backend engineers. Performance and infrastructure engineers. Data engineers. Database and search engineers. Competitive programmers. Algorithm enthusiasts. Compiler and runtime enthusiasts. Compression and codec developers. Open-source contributors. Professionals. Anyone interested in performance engineering, from any country. Team Size 1–4 members. Solo participation is allowed. Cross-college teams are allowed. Cross-disciplinary teams are allowed. Participants may enter as many tracks as they like. At least three completed tracks are required to qualify for the overall Speed Demon title. Event Format The hackathon runs entirely online for 72 hours, from October 30 to November 2, 2026. Every track provides a fixed task, a dataset, and a reference implementation. Participants must develop an optimized solution that produces exactly the expected output. During the event, teams are expected to: Study the published specifications and reference implementations. Design and implement optimized solutions for their chosen tracks. Verify correctness against sample datasets and expected SHA-256 hashes. Profile their implementations and measure performance improvements. Build Docker images that work without network access. Run their implementations within the specified CPU and memory limits. Submit source code, Dockerfiles, and the required documentation. Maintain a PERF.md optimization log documenting experiments and measured results. Monitor the sample leaderboard during the competition. Prepare their final repositories for automated evaluation. Final scoring uses holdout datasets generated from the same distributions with previously unseen seeds. This prevents participants from winning by hardcoding answers or overfitting to known input data. Five Tracks There are five tracks, each targeting a different engineering challenge. Participants can enter as many as they like. Track A - Ingest: Aggregate 3.1 million structured log lines into per-endpoint request counts, p50/p95/p99 latency percentiles, and hourly error rates. The reference implementation takes 41.2 seconds and uses 2,940 MB of memory. Track B - Index: Build an inverted index over a 250 MB text corpus and answer 50,000 Boolean and phrase queries. Build time and query time are measured separately and combined for scoring. Track C - Join: Parse a 204 MB hostile CSV dataset, handle complex encoding and quoting edge cases, join it against a lookup table, and emit sorted aggregates. Exact correctness is mandatory. Track D - Traverse: Process a directed, weighted graph containing 5 million edges and answer 2,000 shortest-path queries. Preprocessing is allowed and included in the timed run. Track E - Compress: Implement lossless compression for a fixed 100 MB corpus containing text, logs, JSON, and binary data. The score balances compression ratio and decompression throughput, with compressed size replacing peak memory in the efficiency coefficient. Every track has a published specification, dataset format, reference implementation, timeout, and technical restrictions. General-purpose libraries that solve the entire task are prohibited where specified in the relevant track rules. Efficiency Coefficient The efficiency coefficient combines five factors, each measured against the reference implementation: Wall Time (T) - 40%: Execution time measured by the official benchmarking harness. Peak Memory (M) - 20%: Maximum resident memory recorded by the execution environment. For Track E, compressed size replaces peak memory. Source Size (S) - 15%: Normalized source-code size after canonical formatting and specified transformations. Dependency Weight (D) - 15%: Cost associated with non-standard-library runtime dependencies and vendored source. Scaling (X) - 10%: Performance as input size increases. The efficiency coefficient is calculated using a weighted geometric mean: K = (Tref/T)^0.40 × (Mref/M)^0.20 × (Sref/S)^0.15 × (Dref/D)^0.15 × (Xref/X)^0.10 Individual factors are capped at 50×. The coefficient is converted into a 0–100 speed score using a logarithmic scale: Speed Score = 100 × ln(K) / ln(25) The result is clamped to the range 0–100. A coefficient of 25 or higher earns 100 points, a coefficient of 5 earns 50 points, and a coefficient below 1 earns zero. The final score for each track consists of: 65% Machine-Measured Performance: The speed score derived from the efficiency coefficient. 35% Human Judging: Code quality, optimization technique, and the required performance log. The overall Speed Demon title requires at least three completed tracks and is determined using the geometric mean of the track totals. This rewards consistent performance across multiple workloads rather than allowing one exceptional result to compensate for weak results elsewhere. What Counts as Done Your submission must build and execute successfully through the official Docker-based benchmarking harness. The implementation must: Read input from the designated /data directory. Write the required output file to /out. Exit successfully with code 0. Build without network access. Run within the published timeout and memory limits. Produce output whose SHA-256 hash matches the expected hash exactly. Respect the sandbox restrictions on networking, filesystem access, and dataset modification. The official harness executes each submission five times. The highest and lowest timings are discarded, and the mean of the remaining three runs determines the wall-time measurement. Self-reported timings are not accepted. The official execution environment determines the results. AI Tools AI coding assistants are allowed and expected, including Claude Code, Cursor, Codex, GitHub Copilot, local AI models, and others. AI usage itself is not scored. We evaluate whether the final implementation works, whether the optimization decisions are technically sound, and whether the team can explain and defend the software it built. Dependencies, frameworks, and development tools are permitted, subject to track-specific restrictions. Dependencies contribute to the dependency-weight calculation, and libraries that solve the entire task are prohibited where specified. What You Need to Submit Each team must submit: Public GitHub repository. OSI-approved open-source license. Working implementation for each entered track. Dockerfile that builds offline and satisfies the execution contract. Required source code and build scripts. PERF.md documenting optimization experiments and measured results. README.md explaining the approach, dependencies, and future improvements. .speed-demon.toml containing the required machine-readable submission metadata. Each track submission must include its own directory and the files required by the official specification. Your PERF.md should explain the optimization journey, including the original baseline, each significant change, measured improvements, unsuccessful experiments, and the reasoning behind the final implementation. A demonstration video of under three minutes is optional. It is not scored, but strong demonstrations may be included in the event's results write-up and archive. Judging Criteria Projects will be evaluated on: Machine-Measured Performance - 65%: The speed score derived from the published efficiency coefficient and official benchmark results. Code Quality - 15%: Readability, maintainability, error handling, and clarity of implementation decisions. Optimization Technique - 12%: Algorithm selection, data structures, memory layout, and the quality of performance optimizations. Optimization Log - 8%: The quality of PERF.md, including measured improvements, unsuccessful experiments, and technical reasoning. Judges review submissions independently without seeing the measured leaderboard before submitting their evaluations. This keeps technical review separate from numerical performance results. Prizes $2,000 Total Prize Pool Speed Demon - Overall Champion: $800 Runner-up: $400 Third Place: $200 Highest Efficiency Coefficient: $200 The Lean Award: $100 Write-Up Side Quest: 3 winners × $100 = $300 The overall champion is determined by the highest geometric mean across at least three tracks. The Highest Efficiency Coefficient award goes to the submission with the largest recorded coefficient on any track, regardless of whether the team qualifies for the overall title. The Lean Award recognizes the best efficiency coefficient achieved with the smallest normalized source. The Write-Up Side Quest rewards insightful technical accounts of the optimization process, including bottlenecks, experiments, unexpected regressions, and measured improvements. Winners receive $100 each. Rules Teams must contain 1–4 members. Solo participation is allowed. Any programming language is eligible if it builds offline in a Linux container and runs on x86-64. All project code must be written during the official 72-hour hackathon window. Planning, research, documentation reading, algorithm design, and environment preparation are allowed before kickoff. No project code may be committed before kickoff. AI coding assistants and development tools are allowed. Dependencies are allowed but contribute to the dependency-weight calculation. Libraries that solve the entire task are prohibited where specified in the track rules. The output must match the reference output exactly. Precomputed answers, cached solutions, and input-specific shortcuts designed to bypass the task are prohibited. The Docker image must build without network access. The submission must respect the official CPU, memory, timeout, and sandbox restrictions. The GitHub repository must be public at code freeze. The repository must use an OSI-approved open-source license. The required PERF.md and .speed-demon.toml must be included. The official benchmarking harness determines the final measurements. At least three completed tracks are required for the overall Speed Demon title. The scoring formula and rules are frozen before kickoff. Violations of correctness, sandbox, or submission requirements may result in a zero score or disqualification under the official rules. Important Dates All times are in UTC. October 2, 2026: Registration opens. October 26, 2026: Team formation closes. October 28, 2026 at 18:00 UTC: Track specifications, reference implementations, and the complete scoring harness are published. October 30, 2026 at 06:00 UTC: Sample datasets and expected hashes are released. October 30, 2026 at 18:00 UTC: Hackathon begins; full datasets are released and the live leaderboard opens. October 31, 2026: Standings and a technical teardown of the reference implementation are published. November 2, 2026 at 18:00 UTC: Code freeze and submission deadline. November 2–4, 2026: Official scoring runs. November 3–12, 2026: Judging window. November 10, 2026: Write-Up Side Quest closes. November 13, 2026: Winners announced. About Hackathon Raptors Hackathon Raptors is a Community Interest Company that runs online hackathons for working engineers and developers. Its events bring together builders from around the world to solve practical engineering challenges and demonstrate their technical skills through working software. Speed Demon follows that approach by making performance, correctness, and reproducible measurements the centre of the competition. 72 hours. Five tracks. One reference implementation to beat. Every millisecond is a soul.