Skip to content
View uaesivakumar's full-sized avatar
🎯
Focusing
🎯
Focusing

Highlights

  • Pro

Block or report uaesivakumar

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
uaesivakumar/README.md

Sivakumar Chandrasekaran

AI Solutions Architect · Abu Dhabi, UAE 🇦🇪

I architect and ship production AI systems — RAG platforms, multi-agent orchestration, LLM infrastructure — proven across multiple industries, including one of the world's most regulated.

Website LinkedIn Email



👋 About

I design end-to-end AI systems that go beyond demos — governed, auditable, and running in production for real users.

  • 🚀 9 AI systems designed and shipped, 5 live today — across 8 domains: talent intelligence, sales intelligence, consulting & decision support, financial services, EdTech, civic analytics, HR & performance, and data engineering
  • 🏗️ 20 years of experience: 8 years in technology delivery (100+ client projects) followed by 12 years in enterprise environments, including Emirates NBD where I was recognised as a top 1% performer
  • 🛡️ Regulated-industry depth: I understand compliance, auditability and governance from the inside — skills that transfer to any enterprise AI deployment
  • 📜 5 provisional patents filed for novel AI architectures

🔒 A note on links & source code: my products are commercial and client-facing, so I keep source code and direct product links private here out of confidentiality. Full case studies and live demos are available at sivakumar.ai or on request.


🧩 Open Source

Repository What it is
llm-gateway Multi-provider LLM routing — automatic failover, circuit breaking, a bounded latency budget, and cost tracking that refuses to guess. One runtime dependency, 115 offline tests, MIT.
ai-architecture-notes Design decisions from production AI systems with the trade-offs left in — agent governance, auditable LLM decisions, hybrid retrieval, and why AI programmes actually fail.
ai-systems-portfolio Case studies of all nine production systems across six industries — architecture, decisions and trade-offs, with source kept private.

🚀 Featured Systems

System Domain What it does Status
Inteller Talent intelligence Ghost-job detection & resume tailoring across 30+ job platforms 🟢 Live
SKC Digital Consulting / decision support Executive simulation built on 30+ documented AI programme failures 🟢 Live
RM Assistant Financial services Relationship-management copilot with structured memory persistence 🟢 Live
AI Leads Portal Sales intelligence Conversational intent capture & pipeline scoring with self-learning feedback loops 🟢 Live
PremiumRadar B2B sales intelligence Governed intelligence platform for regulated industries — 2,600+ source files, 38+ services, 5 provisional patents 🤝 Design-partner stage
Payroll Enrichment Data engineering ADGM registry extraction and enrichment pipeline 🟢 Live
Arsha LMS · Chunav · Coach EdTech · Civic analytics · HR Multi-tenant learning platform, election analytics, performance evaluation 🛠️ In design / build

🏗️ Architecture Highlights

  • Multi-provider LLM gateway — failover across Anthropic, OpenAI and Vertex AI with per-request cost tracking
  • Governance by design — the reasoning layer holds no credential that can change anything; deterministic state machines with 6 decision gates
  • 11-layer discovery pipeline for entity resolution and enrichment
  • Retrieval done properly — pgvector + Neo4j hybrid search, structured memory, full audit trails

🧠 What I Can Prove

Claims are cheap. Each capability below links to the code or the write-up that demonstrates it.

Capability Evidence
LLM platform engineering — multi-provider routing, failover, circuit breaking, cost governance llm-gateway — working code, 115 tests, 94% coverage
Agent governance — reasoning layers that hold no credential able to change state The reasoning layer holds no credential
Auditable AI decisions — deterministic gates around a probabilistic core Deterministic gates
Hybrid retrieval — pgvector for passages, graph for identity and relationships What vectors miss
AI programme diagnosis — 30+ documented failures and where they actually go wrong Why AI programmes fail
Regulated-industry delivery — compliance, auditability and governance from the inside 12 years in banking · 5 systems live in production

Day to day: Python · FastAPI · Node.js · Next.js · PostgreSQL / pgvector · Neo4j · Firestore · Redis · GCP (Cloud Run, Pub/Sub) · Docker · Anthropic / OpenAI / Vertex AI


📊 GitHub Activity

Contribution graph

Most commits land in private product repos — the graph reflects private contribution counts.


🤝 Open To

AI Solutions Architecture · AI / ML Engineering · LLM Platform Engineering · Technical Leadership

Across enterprise SaaS, fintech, government, consulting and product companies — Abu Dhabi / UAE (on-site or hybrid), and remote worldwide.

Pinned Loading

  1. uaesivakumar uaesivakumar Public

    AI Solutions Architect · Abu Dhabi — production AI systems: RAG, multi-agent orchestration, LLM platforms. 9 systems shipped across 8 domains, 5 live.

  2. ai-systems-portfolio ai-systems-portfolio Public

    Case studies of 9 production AI systems across 8 domains — architecture, decisions and trade-offs. Source private; the reasoning is public.

  3. llm-gateway llm-gateway Public

    One interface across Anthropic, OpenAI and Google (Gemini API and Vertex AI) — automatic failover, circuit breaking, bounded latency and per-request cost tracking. One dependency, no SDK lock-in.

    Python

  4. ai-architecture-notes ai-architecture-notes Public

    Design decisions from building production AI systems in regulated environments — agent governance, auditable LLM decisions, hybrid retrieval, and why AI programmes actually fail.