PathOS
Your resume --> Any Codebase. We'll show you how.
Created on 15th February 2026
•
PathOS
Your resume --> Any Codebase. We'll show you how.
The problem PathOS solves
The problem it solves
India graduates 1.5M+ engineers annually, mostly from tier-2/3 colleges. They have talent but lack a structured path from academic theory to production codebases. Students jump from YouTube tutorials to complex repos and get lost—they don't know what prerequisites they're missing.
We solve this by:
- Analyzing any GitHub repo against a user's resume to identify skill gaps
- Generating personalized learning paths using knowledge graphs of real codebases
- Enabling progressive skill-building through actual code, not toy tutorials
Use cases:
- Students: Bridge the college-to-industry gap with clear learning roadmaps
- Companies: Faster onboarding—auto-generate prerequisite maps for internal repos
- Universities: Modernize curriculum by mapping textbook concepts to real-world code
We're building the missing infrastructure between India's talent and its tech industry.
Challenges we ran into
Challenges I ran into
Challenge 1: Analyzing Large Codebases at Scale
Naively exploring large repositories is computationally expensive. We built an AST-based knowledge graph engine that indexes codebases into nodes (classes, functions, modules) and maps their relationships. This enables semantic search over the graph instead of brute-force text scanning.
What made this hard: No off-the-shelf solution exists. We studied multiple open-source implementations (tree-sitter, language servers, static analysis tools) and adapted patterns to work across multiple languages in 7 hours.
Challenge 2: Skill Gap Analysis with Fuzzy Inputs
Resumes are unstructured ("JavaScript" vs "React ecosystem" vs "full-stack web dev"), and repo requirements are implicit. We built a pipeline that:
- Extracts skills from resumes using LLM
- Identifies prerequisite concepts from the knowledge graph
- Fuzzy-matches and scores to compute readiness
What made this hard: Generating structured, low-variance outputs from noisy inputs required aggressive prompt engineering and JSON schema validation.
Challenge 3: Generating Logical Learning Paths
It's not enough to identify gaps—we needed to recommend specific concepts and repos in the right order. We rank prerequisite concepts by dependency (e.g., "learn async before distributed systems") and suggest progressively harder repositories from our indexed knowledge base.
What made this hard: Building a concept dependency graph in 7 hours meant hardcoding heuristics instead of training a model. We prioritized a working demo over perfection.
Tracks Applied (1)