SYSTEM SPECIFICATION

Serverless AST Intelligence & Multi-Model Pipeline

DevContext leverages a serverless architecture combining Tree-sitter AST parsing, multi-model AI routing, and progressive streaming to deliver verified codebase intelligence in under 30 seconds.

Topology Overview

AWS Serverless / Bedrock

Component Breakdown

STAGE 0

Ingestion & AST Mapping

Clones and normalizes the target GitHub repository. Tree-sitter isolates developer-authored code from vendor dependencies (such as node_modules or standard framework templates), producing a clean AST context map cached in S3.

STAGE 1 & 2

Parallel Agent Synthesis

To achieve a sub-30 second latency SLA, specialized Lambda workers inspect architecture, concurrency patterns, and storage bottlenecks in parallel. A reducer agent verifies citations before saving state to DynamoDB.

STAGE 3

Real-Time Mock Interview

WebSocket connection tracks live interview sessions. Questions adapt to the candidate responses, scoring clarity, depth, and communication against verified code call graphs in real time.

Verification & Grounding Evidence

Verified Trace
Claim: Zero Hallucination Line ReferencingAST Match: 100%
$ devcontext query --repo /src/auth --question "Does it use bcrypt?"
[+] Searching abstract syntax trees across 142 repository files...
[+] AST isolation: excluded 12,400 vendor files (node_modules, /dist)
[+] Grounding engine verified lines 45-89 in auth/hash.go
[✓] ANSWER: Yes. Evidence: `bcrypt.GenerateFromPassword()`
[✓] CONFIDENCE: 100% (Direct AST reference matched)