Building systems that matter
Passionate about shipping impactful products, scaling teams, and solving hard problems in high-throughput messaging and distributed systems — and building with LLMs, RAG, agentic AI, Reasoning + Acting, and MCP.
About
01Passionate about building systems that scale, leading high-performing teams, and creating lasting impact through excellence in design and execution. Over the years I have built and operated large-scale distributed messaging and SaaS communication platforms — across Email, SMS, Push, and real-time systems — handling multi-million-recipient workloads with reliability at their core. Alongside that, I build AI agents and RAG systems as hands-on projects in Go — Reasoning + Acting agent loops, MCP-based tooling, LLMs (Claude and local Ollama), and retrieval-augmented generation over multiple vector stores (ChromaDB, Qdrant, Pinecone).
Recent focus — AI agents, RAG & agentic architecture, with new engineering writing shipping regularly.
Roles & Perspectives
Experience
02Self-employed
Freelance Software Engineer
Jun 2026 - Present
Building high-concurrency autonomous AI agents and complex RAG pipelines in Go, using vector and document databases (Chroma, Pinecone, Qdrant, MongoDB) for semantic retrieval. Architecting low-latency streaming voice loops (STT/TTS) over WebSockets and Redis Streams with sub-100ms performance. Optimizing local LLM runtimes, caching layers, and token streaming, while managing end-to-end product delivery and Docker deployments for diverse clients.
Brevo (formerly Sendinblue)
Sr. Lead Engineer / Engineering Manager
2015 - May 2026
Built and operated large-scale messaging and SaaS communication platforms supporting millions of recipients. Led backend engineering teams across platform modernization, distributed systems, and reliability initiatives. Architected asynchronous pipelines, event-driven systems, and observability infrastructure. Mentored engineers on scalable design and production excellence.
Intecons Software Lab
Programmer Scale II / Programmer Scale I
2012 - 2014
Built backend components and data workflows across diverse domains—freight, logistics, geo-tracking, property, and social platforms. Developed strong fundamentals in system design and production support.
Highlights
03Re-architected email at scale
Migrated email sending from PHP to Go and moved it onto an event-driven producer/consumer pipeline — scaling the platform to process 350M+ emails per day with horizontal scalability and headroom for peak traffic.
Contact-level Kafka fan-out
Re-architected a marketing SMS pipeline from campaign-level to contact-level processing (~28 → ~600 msg/s), cutting a 500K-recipient campaign from hours to under 5 minutes — with exactly-once processing and sub-5-minute feature-flag rollback.
AI-native engineering
Build with agentic AI and retrieval — shipped a Reasoning + Acting agent and a multi-backend RAG system — and ship production architecture with an AI agent harness (CONTEXT-driven planning, verifier gates, atomic commits), reaching ~3× delivery throughput versus a solo baseline.
Engineering leadership
As Tech Lead / Engineering Manager across messaging-platform squads, authored RFCs/ADRs before implementation, scoped epics into shippable work, mentored engineers, and owned feature-flag release governance.
Architecture & Scale
- Architected and led the build of an entirely new messaging channel — Kafka topic design, 3-layer retry (in-process → cloud tasks → DLQ), template rendering, pricing, and observability — from infra provisioning to production.
- Ran a zero-downtime sync→async migration via staged feature-flag rollout, then retired the legacy path across services.
- Standardized context propagation across 7 microservices; led a multi-cluster Kubernetes migration with ArgoCD lifecycle management and cluster decommissioning.
- Applied a dual-write pattern for zero-downtime data-store migrations.
- Authored end-to-end RFCs/ADRs for a compliance subsystem (sender-ID approval in regulated markets) — architecture, API contracts, and a single-source-of-truth data model.
Reliability & Operations
- Resolved a platform-wide outage — every authenticated route failing — by root-causing a middleware-ordering bug, then adding tracing to prevent recurrence.
- Built observability: per-call tracing, structured logging, top-client dashboards, and APM noise reduction (business outcomes ≠ system errors).
- Shipped anti-abuse controls that stopped carrier-level domain blocking and protected sender reputation in key markets.
- Drove 10+ security-vulnerability remediations across Go and Node.js, plus a multi-year initiative to reduce L3 support toil.
- Built multi-carrier routing resilience across five carriers and a multi-channel platform (SMS, RCS, push), with live-monitored staged rollouts.
Leadership & Delivery
- Tech Lead / Engineering Manager — managed 8+ engineers (incl. tech leads and senior ICs) across two engineering tribes; technically led a team building a new channel from scratch.
- Owned hiring, onboarding, performance reviews, and career growth for backend engineers — mentoring ICs toward senior scope.
- Authored RFCs/ADRs before implementation, scoped epics into shippable work, and ran sprint planning with consistent over-delivery (one quarter at ~140% of committed scope).
- Acted as technical escalation lead for production incidents — owning triage, mitigation, and follow-up across the messaging platform.
- Treated code review as first-class tracked work and championed team recognition.
- Unblocked an enterprise deal with a backward-compatible public-API extension and cross-team SDK coordination.
- Wrote platform documentation and end-to-end system maps that reduced key-person risk and accelerated cross-team onboarding.
Skills & Expertise
04Languages & APIs
AI Engineering
Distributed Systems & Messaging
Cloud & Infrastructure
Data & Storage
Reliability & Observability
Leadership & Delivery
Projects
05
Voice AI Support Agent v2
A production-grade, low-latency, multilingual Voice AI banking agent in Go. Dynamic English/Hindi Turn Supervisor, speculative cache probing to bypass LLM latency, WebRTC (LiveKit) gateway, and Cassandra audit logs.
Agentic AI (Python)
A full Python rebuild of the agent — FastAPI services (agent · retrieval · MCP) behind a python.org-styled UI. Conversational memory, native PDF retrieval over Pinecone, live web search with clickable citations, and clarify-back.
Omni-RAG
A multi-backend RAG system spanning ChromaDB, Qdrant, and Pinecone with keyword-sensitive filtering and semantic fallback — resolving the short-query vector bias that breaks out-of-the-box RAG on real-world data.
Agentic AI
A Go-based agentic AI system using a Reasoning + Acting loop, tool-calling, and MCP integration. Combines Tavily search and Ollama-backed local models for autonomous multi-step task execution.
Rejected.AI
Work in ProgressExploring early-stage ideas and research in AI. A living lab for experimenting with novel approaches to agent behavior, learning systems, and edge cases.
Agentic AI (Python) — Reasoning + Acting engine
A full Python rebuild of the agent: FastAPI services (agent · retrieval · MCP) behind a python.org-styled UI. Same Reasoning + Acting loop — conversational memory, native PDF retrieval over Pinecone, live web search with clickable citations, and clarify-back — re-implemented end-to-end in Python.
Agentic AI (Go) v2 — Your Library Knowledge Base
The agent grew a memory and a library: conversational retrieval with per-source provenance, native PDF search over Pinecone with a books catalog and clarify-back, live web search with clickable citations that persist, and PDF ingestion — all behind Web / My Library toggles.
Agentic AI (Go) v1 — Reasoning + Acting engine
A Go-based agent that thinks step-by-step: a Reasoning + Acting loop interleaves thinking with tool calls across PDF retrieval and live web search, grounding multi-step answers in real sources — built with MCP tooling, Tavily search, and Ollama-backed local models.
Voice AI Support Agent v2 — Multilingual Banking Edge
A high-performance conversational voice edge system. Dual Branch A (Speculative prefill) and Branch B (Qdrant semantic cache probe) run in parallel to halt LLM execution early, saving up to 80% of GPU cycles. Integrated with LiveKit WebRTC, Redis session state, MongoDB transaction ledgers, and a Cassandra time-series log sink.
Omni-RAG — end-to-end retrieval-augmented generation
A full RAG pipeline in action: ingest a PDF — secured to GridFS, layout-extracted, embedded, and upserted to Pinecone with live SSE progress — then ask questions and get grounded, cited answers via semantic Top-3 vector retrieval and a streamed LLM.
Education
06B.Tech (Computer Science)
Dr. A.P.J. Abdul Kalam Technical University
Writing
07Let's build something reliable together.
Open to conversations about distributed systems, AI engineering, and engineering leadership — whether it's a role, a collaboration, or just trading ideas.