public AI systems
Open code, inspectable architecture
Available for select AI systems work
Dhrubojyoti Gangopadhyay · Kolkata, India
Enterprise instincts.
AI systems that ship.
I design production AI agents, retrieval systems, voice workflows, and operational intelligence for complex businesses.
Kolkata / IndiaWorking worldwide
Agent architecture
DB-tier logic
years in enterprise
Commercial operations at national scale
voice pipeline
Speech to structured CRM output
faster decisions
Database-led operating intelligence
01 / Point of view
AI is useful only when it understands the business beneath it.
I spent 17 years inside pricing, distribution, supply chain, and revenue operations before building AI systems. That field experience changes the architecture: fewer demos, better constraints, clearer ownership, and software designed around measurable outcomes.
02 / Selected work
Systems built for the real world.
Four examples across compliance, orchestration, enterprise tooling, and realtime voice. Each repository opens to the implementation.
Featured system / 01
ComplianceGraph — Agentic Compliance Intelligence
Graph-minded compliance system for tracing obligations, controls, evidence, and reasoning paths instead of treating compliance as flat document search.
- System
- compliance entity graph → retrieval layer → reasoning agent → evidence-backed answer path
- Signal
- shows Dhrubo can translate business risk into AI system architecture
Featured system / 02
SENTRA-AI — Operational Agent System
Current-generation AI operations build focused on agentic execution, workflow orchestration, and practical system behavior rather than static chat.
- System
- task intake → agent plan → workflow execution → telemetry-ready outputs
- Signal
- designed for practical operators, not just prompt demos
Featured system / 03
NEXUS Enterprise AI Operations Platform
True MCP abstraction layer — RAG, GitHub, Filesystem, Browser, Voice, Postgres as discoverable, auditable, executable tools via stdio/SSE JSON-RPC.
- System
- Stateful LangGraph supervisor → MCP tool registry → dynamic routing → parent-doc memory checkpointing → audit logs
- Signal
- Gemini embeddings + lazy-loaded Playwright for resource-constrained hosts
Featured system / 04
Zoya Voice CRM Integration Agent
Orchestrates Vapi to manage WebRTC/SIP telephony, bypassing raw audio socket streams. Parses speech → structured CRM schemas at sub-250ms latency under concurrent high-throughput load.
- System
- Latency: <250ms speech-to-structured CRM schema, zero UI-blocking
- Signal
- handles high-throughput inbound calls without blocking
03 / Operating method
From ambiguity
to dependable action.
The cinematic part is the clarity: watching a messy operating problem become a system people can trust.
Understand
Start with the operating reality: people, incentives, exceptions, and risk.
Architect
Choose the right memory, tool, model, schema, and control boundaries.
Ship
Build the smallest production system that can survive real usage.
Observe
Trace outcomes, catch failure modes, and improve from evidence.
04 / Experience
Operator first.
Builder by conviction.
A career spanning enterprise commercial leadership and hands-on AI engineering, connected by the same obsession: making complex systems work better.
Jan 2025 — Present
Self-Employed · Remote
Independent AI Engineer
Shipped 31 public AI systems — stateful graphs, advanced RAG networks, custom agent tracing backends, low-latency API wrappers, compliance intelligence, and agentic workflows
2007 — 2024
ACC Ltd · Lafarge · MP Birla Group
AGM & Senior Manager — Revenue Operations
Directed commercial operations, designed distribution databases — reduced decision latency 40%
2005 — 2007
University of Kalyani
MBA — First Class
Specialization in Operations Management & Marketing Strategy
Agentic & Graph RAG & Protocols Voice & Realtime Web Automation Databases & Backend Frontend & LLMs Agentic & Graph RAG & Protocols Voice & Realtime Web Automation Databases & Backend Frontend & LLMs
05 / Contact
Have a hard AI problem?
Tell me what is slow, manual, unreliable, or impossible today. I'll tell you where an AI system can create real leverage.