People who need an answer cannot get it, because the answer lives behind SQL, five admin screens, or someone else’s calendar.
Production RAG, GraphRAG, agent orchestration, and LLM applications that run against live business data.
Multi-Agent Conversational AI →The knowledge exists, but it is spread across thousands of unstructured documents with no schema and no shared vocabulary.
Large-scale document processing, entity and relationship extraction, graph-based retrieval, and vector search.
GraphRAG Knowledge Engine →A system the whole company runs on, all day, where an hour of downtime means shipments stop.
Cloud architecture, distributed systems, APIs, data pipelines, and the observability to know when they degrade.
Secure Logistics ERP →An AI system that can read the database can read the wrong tenant’s data. The access model has to be part of the design.
Permission-aware generation, post-generation query rewriting, read-only validation, and enterprise access control.
Permission rewrite on generated SQL →Production systems across knowledge graphs, multi-agent AI, FinOps, search, logistics, and analytics — from 0-to-1 through enterprise scale.
Full GraphRAG pipeline — ingests documents, builds knowledge graphs with entities, relationships, topics, and communities via vector similarity. 17 activity classes, 14 prompt templates.
Read the case study →Supervisor pattern routing to 5 specialized agents — Reporting, Admin, System Setup, Knowledge, and Support Tickets — across ~30 registry-driven tools. Natural-language SQL with permission rewrite, seven-collection RAG, 100+ report types, SignalR streaming of the reasoning trace.
Read the case study →Automatically reviews pull requests with inline comments and star ratings. 10+ languages, penalty-based scoring, fix verification mode.
View More →Nightly FinOps agent: 19 collectors across 3 phases, 13 anomaly detectors, 4-tier breach thresholds. 20+ standing questions answered deterministically; an LLM handles only the reasoning. Unused resources, query cost attribution, and rightsizing — 30% cost reduction.
Read the case study →Composite health score (0–100) per service. Rolling 30-day baselines for regression detection. Nightly collection of Azure Monitor metrics and SQL DMV telemetry sampled at 15-second intervals, 13 anomaly checks, 4-tier breach thresholds — surfacing degradation before impact.
View More →Synchronizing records into Azure AI Search. Bulk indexing via Durable Functions, real-time incremental sync via Service Bus.
View More →End-to-end ERP — Finance, Operations, Booking, Hub Management, CRM, Route Planning. 2,500 daily users, 70K dockets/month.
View More →Capturing clicks, mouse movements, text input, navigation and URL transitions to reconstruct actual user workflows. Built the entire stack from scratch.
View More →Tools I build in public — agent kits, package managers, review pipelines, libraries, and UI plugins other engineers actually use.
Scaffold an agent on Apra Fleet. The shell already runs. /agent-builder in Claude walks through the spec, the plan, and the tools that are specific to the job.
npm ↗$ npm create @apralabs/agent-kit (opens in a new tab) Discover, install, and manage reusable Skills, Commands, and Agents from GitHub-hosted registries — through an interactive TUI or the CLI.
npm ↗$ npm install -g plugvault (opens in a new tab) Runs in CI, produces structured findings, scores code quality, and posts inline comments to Azure DevOps or GitHub.
Docker Hub ↗$ docker pull dsiddharth2/codehawk (opens in a new tab) Composer package for reading barcodes and QR codes. 277K+ Packagist installs, 133 GitHub stars, and featured on the upstream ZXing repo.
Packagist ↗ 277K+ installs$ composer require dsiddharth2/php-zxing (opens in a new tab) Small MooTools plugin that packs tiles into a Pinterest-style column grid — including infinite-scroll galleries. Inspired by Wookmark for jQuery.
Quick answers about my work, experience and stack. Anything else, email me.
dsiddharth2@gmail.comI work the full path from ambiguous problem to production: understanding the problem, designing the system, building it, deploying it, and iterating once people are using it.
Apra Labs · Architect · previously Associate Architect · Aug 2019 – Present MildlyClassic · Engineering Team Lead · May 2016 – May 2019 UXgage · Co-Founder & CTO · Oct 2015 – Apr 2016 Firaa.me · Side Project · Built at IdeaFlask · Sep 2014 – Sep 2015 Samvit Learning · Founder · Apr 2011 – Apr 2016 dotCORD IT Solutions · Web Developer Intern · 2010–2013
Built the AI and data platform layer for a unified facility management product serving 300+ US facilities.
Multi-Agent Conversational AI — Ask a building a question in plain English — five specialist agents over a 4 TB operations database, instead of five admin screens. GraphRAG Knowledge Engine — Turning 20K unstructured support tickets into an answerable knowledge base. Cloud Cost Intelligence Agent — Finding out why the cloud bill moved — every night, before anyone has to ask. Identified 30% savings across the Azure estate.
Multi-Agent Conversational AI →GraphRAG Knowledge Engine →Cloud Cost Intelligence Agent →
Apra Agent Kit — A starting point for a new agent Plug — Package manager for Claude Code skills CodeHawk — AI-powered pull request review pipeline PHP ZXing — PHP wrapper for the ZXing barcode library MooPin — Pinterest-style column grid for MooTools
Languages & Frameworks: Python · TypeScript · C# / .NET 8, PHP · React · Node.js, LangChain · LangGraph AI & Data: GPT / LLM integration, FAISS · Vector search, GraphRAG · Multi-agent systems Infrastructure: Azure (Durable Functions, Service Bus, Cosmos DB), AWS · Azure AI Search, Neo4j · Cassandra Practices: System design & architecture, CI/CD · DevOps · FinOps, Team hiring & mentorship
Email dsiddharth2@gmail.com. Always open to interesting problems and collaborations.