AI Solutions Architect · Presales & Solution Engineering
Abu Dhabi, UAE
RAG platforms, multi-agent orchestration, LLM infrastructure — and 12 years on the customer's side of the table.
That combination is the point. Most people in this field are one or the other.
0
AI systems shipped, across 8 domains
0
years on the customer's side of the table
0
years across engineering & the commercial front line
Top 0%
performer, Emirates NBD
Scroll to explore
At a glance
9
Systems designed and shipped
Across 8 domains, from talent intelligence to civic analytics
5
Running live today
inteller.ai, skc.digital and three internal tools
3
LLM providers behind one gateway
Anthropic, OpenAI and Vertex, with failover
12 yrs
On the customer's side of the table
Emirates NBD — regulated-industry depth
Background
Engineers are easy to hire. Bankers are easy to hire. The scarce profile is someone who has done both long enough to be trusted in either room.
Sivakumar Chandrasekaran
Abu Dhabi, UAE
2006 — 2014
B.Tech in Information Technology, then eight years delivering web and digital projects end to end. 100+ client sites, full lifecycle — requirements, build, launch, support. I learned that shipping is a different skill from knowing.
2014 — Present
Twelve years at Emirates NBD — ranked among the top ten most AI-mature banks in the Middle East and Africa. Retail banking, B2B sales, priority relationships. The GEMS Sapphire Award recognised me as a top 1% performer across the bank, against 150%+ target achievement, alongside a Certificate of Achievement from the Group CEO. Twelve years of discovery conversations and carrying a number, and a clear view of which processes are expensive, which data is trustworthy, and which problems are worth solving.
2023 — Present
Nine systems designed and shipped across eight domains, five of them running today — RAG architectures, multi-agent orchestration, multi-provider LLM gateways, vector search, all on GCP. Not prototypes: systems with migrations, audit trails and monitoring. The regulated-industry work is the deepest, and it transfers.
The banking years are not a detour from the engineering. They are the reason the engineering is worth anything.
How I Build
Everything below is running in a repository I own, not a course I finished. Happy to walk through any of it, or demo it live.
01
One interface across Anthropic, OpenAI and Google (Gemini API and Vertex AI), with automatic failover between them, circuit breaking, prompt template management, structured outputs and per-request cost tracking. An MCP provider adapter keeps tools portable across models. Extracted and open-sourced as llm-gateway — MIT, 115 tests.
02
Purpose-built agents rather than one prompt doing everything — research, compose, compliance, deliverability and learning agents, coordinated behind a next-best-action layer with a fact-checking pass.
03
RAG-first: internal knowledge is searched before any external call. Vector search on pgvector alongside a Neo4j graph for entity relationships, with entity resolution run before retrieval so the query resolves to the right entity rather than to something that merely reads like it. Structured per-customer memory survives across sessions.
04
The reasoning layer holds no credential that can change state — it recommends, and the authority to act lives in a separate service the model cannot reach. State changes pass deterministic gates, not model judgement, and every decision lands in a full audit trail. Plus what auditors actually ask for: export controls, deletion paths, tenant scoping, and a scenario benchmark that tests whether the system is getting better rather than just different.
05
Everything I build, I deploy and keep running. Containers on Cloud Run, CI/CD on Google Cloud, async work on Pub/Sub and cron workers, secrets in Secret Manager, secret scanning wired into CI — tested and auditable, not a notebook that worked once.
06
The surface people actually touch — web apps in Next.js and FastAPI, voice input via Whisper, on-device OCR for a mobile overlay, and D3 for data-heavy views.
The shape of it
Systems
Designed, built, deployed and maintained by me, around a full-time role. Status is stated plainly. I can demo any of them.
Governed sales intelligence for regulated industries. The model recommends; it is structurally unable to act. Multi-agent, RAG-first, four services. Built and ready for go-to-market.
Ghost-job detection, honest fit scoring and authentic resume tailoring across 30+ job platforms. Built and launched in four months.
Executive simulation and judgement frameworks reasoning over 30+ documented AI programme failures.
Personal relationship management for a banker — structured per-customer memory, a follow-up engine and target tracking, so no lead slips.
Conversational intent capture plus the sales pipeline behind it — stages, callbacks, lost reasons, daily task view. Offline-capable for event venues.
ADGM company data pipeline — scraping, cleaning and enrichment at scale.
Multi-tenant learning platform for competitive exams, with AI-generated tests and subjective grading.
Election analytics with D3 visualisations, multi-language support and an all-India data model.
Performance evaluation system designed as microservices.
Eight domains across the nine — talent intelligence, sales intelligence, consulting and decision support, financial services, HR and performance, EdTech, civic analytics and data engineering. Eight data models, eight sets of users, eight regulatory shapes. Several run under regulated-industry constraints, which is depth that transfers rather than a sector I am confined to. Full walkthrough and live demo available on request.
Public repositories
The multi-provider routing layer, extracted and open-sourced. One interface across Anthropic, OpenAI and Google (Gemini API and Vertex AI), with automatic failover, circuit breaking and per-request cost tracking.
Design decisions from building production AI in regulated environments, with the trade-offs left in — agent governance, auditable LLM decisions, hybrid retrieval, and why AI programmes actually fail.
Case studies of all nine systems — architecture, decisions and trade-offs. The source stays private; the reasoning is public.
Commercial Track
Plenty of people are polished in front of a customer. Plenty of people can ship a production system. The two records below belong to the same twenty years.
01
Twelve years at Emirates NBD in retail banking, B2B sales and priority relationships. Recognised as a top 1% performer across the bank with the GEMS Sapphire Award, against 150%+ target achievement, plus a Certificate of Achievement from the Group CEO.
02
Twelve years of discovery conversations, objection handling and stakeholder alignment — reading what a customer actually needs against what they first asked for, and holding a room that contains both the sceptic and the budget.
03
Eight prior years in technology delivery across 100+ client engagements — scoping the work, proposing it, and delivering it end to end. The commercial instinct came first; the engineering was built on top of it.
I can run a technical discovery session, design the architecture, build the proof-of-concept, and present it to a board.
Technical Foundation
The engineering is not a recent pivot. It starts with a B.Tech in Information Technology and eight years of full-lifecycle delivery before the banking years, which is why the AI work reaches production instead of stopping at a demo.
Everything listed here exists in a repository I own and deploy. None of it is a course completion or a familiarity claim — the parts that can be shown publicly are open-sourced, and I am happy to walk through any of the rest live.
I know what “done” means, because I am the one who keeps it running afterwards.
B.Tech in Information Technology
Formal technical education
8 Years — Technology & Digital Delivery
100+ client engagements. Full lifecycle — requirements to delivery.
Data Science & ML
Feature engineering, prediction, forecasting, model deployment & monitoring
Full-Stack AI Development
Next.js, Node.js, Python, PostgreSQL, GCP, LLM integration
Cloud & Production
GCP Cloud Run, Cloud SQL, Pub/Sub, Secret Manager. CI/CD, Docker, Redis.
Open Source
llm-gateway — multi-provider LLM routing. MIT, 115 tests.
Latest Insights
Contact
I'm interested in Solutions Architect and presales roles at AI and cloud companies, Data & AI consulting practices, and AI leadership roles inside financial institutions. If the work needs someone who can design the architecture and then hold the customer conversation about it, I'd like to hear about it.
Based in Abu Dhabi. Open to roles across the UAE, on-site or hybrid.
Quick Message