Share of development time that can go into understanding code.
Bending Spoons track
benderrender
Map any repository, navigate by intent, and understand unfamiliar code in minutes.
Analyze a local repo, prioritize the right nodes, and use local AI via Ollama to get to the best entry point with grounded context.
Full-project graph
Local AI
Grounded summaries
full-repo context
ranked entry points
voice + typed navigation
/workspace/benderrender
Current route
app/page.tsx -> ObsidianGraph.tsx -> VoiceNavigator.tsx
Follow the graph layer into the interaction surface that accepts voice or typed intent.
Project Signal
The slow part is not typing. It is orientation.
benderrender is aimed at the part of engineering work where teams lose time first: understanding the shape of the codebase, choosing a credible entry point, and building context before editing.
Evidence Graph
Signals behind the product thesis
Research-backed plus current product data.
Controlled-study improvement from AI support during development tasks.
Tasks completed faster when a visual code view was available.
Current benderrender flow cuts the initial search surface dramatically.
Product
Understand first. Search later.
Three moves: map, navigate, explain.
1. Map
Turn the repo into a useful map.
Walk the folder, resolve imports, and draw dependencies.
- whole graph visible
- centrality and context
2. Navigate
Ask by intent, not by exact path.
Heuristic shortlist, local model, and neighbor-aware refinement.
- ranked entry points
- inferred route visible
3. Explain
Summarize with context.
Short answers grounded with lines and graph neighbors.
- line citations
- incoming and outgoing neighbors
Multiplier X10
X10 here means less scanning and fewer jumps.
It is not a blanket academic claim of ten-times end-to-end speed. It is a measurable compression chain: less search surface, fewer plausible paths, and less time to the first useful file.
Full repo
Start from the whole graph instead of blind file hunting.
->
Ranked shortlist
Reduce the search space to the most likely entry areas.
->
Connected route
Refine through neighbors that keep structural context.
->
Useful entry
Land on a grounded starting point faster.
Research
Research that supports the thesis.
Comprehension, context, and visual orientation do improve productivity.
Program comprehension is still a bottleneck.
About 58% of development time can be spent on program comprehension.
Xia et al., 2018AI assistance moves real elapsed time.
GitHub Copilot made the experimental group 55.8% faster.
Peng et al., 2023Well-placed context lowers cognitive load.
Accuracy improves and cognitive load drops when information appears at the right time.
Adeli et al., 2020Visual views help with fast overview.
Code Cities reduced completion time in 4 out of 6 tasks.
Galperin et al., 2022