code-perf

Example: Code Performance Optimization

Maximize Python JSON parsing throughput with Open Researcher — from baseline ~5,000 ops/sec to ~50,000+ ops/sec through pure code optimization.

Prerequisites

• Python 3.10+
• No additional dependencies (stdlib only)
• CPU is sufficient
• One AI agent installed: claude (Claude Code), codex, aider, or opencode

Quick Start

# 1. Create project directory with a JSON parser + benchmark
mkdir code-perf && cd code-perf

# Write a baseline recursive descent JSON parser (parser.py) and
# a benchmark script (bench.py) that:
#   - Parses a set of JSON test strings
#   - Measures ops/sec (parse operations per second)
#   - Prints ops_per_sec at the end

# 2. Initialize Open Researcher
pip install open-researcher
open-researcher init --tag code-perf

# 3. Launch autonomous research
open-researcher run --agent claude-code

# Or run headless with a specific goal
open-researcher run --mode headless \
  --goal "Maximize JSON parsing throughput (ops/sec) by optimizing the Python parser implementation with better algorithms, data structures, caching, and code-level optimizations" \
  --max-experiments 20

What the Agent Will Try

• Algorithm optimization (recursive descent vs iterative, state machines)
• Data structure replacements (dict vs OrderedDict, list pre-allocation)
• String processing optimization (avoid repeated slicing, use memoryview)
• Caching strategies (memoization for repeated structures)
• Python-specific tricks (__slots__, local variable access, reduce function calls)
• Regex vs hand-written tokenization
• Memory allocation reduction
• Branch prediction-friendly code patterns

Metrics

• Primary: ops_per_sec (higher is better) — JSON parse operations per second
• Evaluation: Run benchmark with test JSON strings, measure throughput
• Typical baseline: ~5,000 ops/sec (naive recursive descent parser)
• Typical best after ~15 experiments: ~50,000+ ops/sec