Overview
Unifying Product Requirements, Implementation, and Verification
Product knowledge, source-code context, and test assets are often distributed across documents, repositories, and separate engineering tools. This makes engineering traceability, context discovery, and quality analysis time-consuming.
IQP is an enterprise Quality Engineering platform that combines graph-based product knowledge, code intelligence, and AI-assisted retrieval in a workspace-oriented product experience.
Product capabilities
High-Level Platform Features
- Product knowledge ingestion — Converts raw product specification files into structured logical concepts.
- Knowledge and code graph indexing — Constructs multi-layered relationships connecting features, source files, and test files.
- Hybrid context retrieval — Combines vector-based semantics with exact matching for reliable concept query answers.
- MCP integration — Provides a Model Context Protocol server, enabling AI coding agents to retrieve precise code and specification snippets contextually.
- Visual graph exploration — Interactive portal interface showcasing feature dependencies, code references, and test coverage paths.
- Workspace and access management — Secures isolated environments for client codebases, user roles, and team configurations.
- Code and test traceability — Supports traceability between product concepts, implementation context, and related test assets.
- Quality-analysis insights — Provides quality-analysis insights across product, implementation, and test artifacts.
My contributions
Direct Technical Responsibilities
- Backend architecture — Designed service boundaries for workspace state, orchestration, graph operations, code intelligence, and user-facing workflows.
- Retrieval — Implemented scoped context retrieval combining semantic, lexical, and graph-based signals.
- Code intelligence — Developed repository indexing and context-resolution capabilities linking product concepts with related implementation and test assets.
- Asynchronous processing — Built background workflows for ingestion, indexing, and graph-maintenance workloads.
- Product integration — Connected backend capabilities with a React/TypeScript portal and MCP-compatible AI tools.
Architecture principles
Design Patterns & System Decisions
Workspace Portal & AI Clients (MCP)
│
▼
┌─────────────────────────────────────────┐
│ FastAPI Control Plane Gateway │
│ - Workspace boundaries & tenancy │
│ - Query routing & MCP tool server │
└──────────────────┬──────────────────────┘
│
┌─────────┴─────────┐
▼ ▼
┌──────────────────┐ ┌──────────────────┐
│ Hybrid Retrieval │ │ Celery Worker │
│ - Dense Qdrant │ │ - Git Ingestion │
│ - BM25 Lexical │ │ - Code Parser │
│ - Code Graphs │ │ - Graph Builder │
└────────┬─────────┘ └────────┬─────────┘
│ │
└─────────┬───────────┘
▼
PostgreSQL + Redis Cache - Central backend as orchestration and access boundary — Managed workspace boundaries, permissions, and service communication behind a central gateway proxy.
- Separation of ingestion workloads from read/query workloads — Kept compute-heavy indexing and specification extraction isolated from retrieval latency.
- Lightweight MCP access for scoped context retrieval — Enabled AI tools to fetch subgraphs matching active development tasks without overflowing context windows.
- Multi-signal retrieval rather than vector-only search — Integrated dense vector similarity with lexical matching and graph traversal, which improves retrieval precision for specialized engineering terminology.
- Long-running processing isolated in asynchronous workers — Offloaded repository scanning and parsing tasks to background workers with managed retry rules.
Technologies
FastAPIQdrantPostgreSQLRedisCeleryMCPReactTypeScriptDocker Compose
Outcome
Contributed to a working enterprise platform foundation that combined product knowledge, source-code context, AI-assisted retrieval, and quality-analysis capabilities in a single engineering workflow.