Skip to content
View sarathalapad's full-sized avatar

Block or report sarathalapad

Report abuse

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

Report abuse
sarathalapad/README.md

Hi, I'm Sarathkumar K

AI Solutions Architect · Forward Deployed AI Engineer · Agentic AI & Secure GenAI

I design and build production AI systems for government and banking: multi-agent workflows, RAG and tool-using agents that run fully air-gapped on private LLMs or on Azure, AWS and Google Cloud, with the governance regulated environments need.

18+ years in enterprise technology, 8+ in AI/ML and data. I work onsite with client teams from use-case discovery to production.

What I build

  • Agentic AI: LangGraph multi-agent workflows, MCP tool calling, human-in-the-loop approvals, bounded research agents
  • Secure RAG: hybrid retrieval (BM25 + dense + RRF), ACL-aware retrieval, grounded answers with citations, evaluation with Ragas / Giskard
  • Sovereign / air-gapped AI: Ollama and vLLM serving, Llama / Qwen, BGE-M3 embeddings, offline model transfer, on-prem OpenShift
  • AI governance: guardrails, PII masking, maker-checker for risky actions, tamper-evident audit, NIST AI RMF, Zero Trust
  • LLMOps and platform: evaluation gates, LoRA fine-tuning, OpenTelemetry tracing, Kubernetes / OpenShift, HA/DR, CI/CD

Open-source projects

Agents, RAG and governance

Project What it shows
governed-rag-agent Auditable AI agent: hybrid RAG, prompt-injection guardrails, PII masking, human approval for risky tools, hash-chained audit, CI eval gates
multi-agent-research-crew LangGraph multi-agent system: planner, parallel researchers with bounded tool use, writer, reviewer loop, human-in-the-loop interrupt
governed-mcp-server MCP server giving agents governed access to IT ops: policy engine, maker-checker approvals, dry-run, rate limits, lockdown, audit
rag-eval-harness RAG evaluation and CI regression gate: hit@k, MRR, nDCG, F1, faithfulness, citation validity, abstention, bootstrap CIs

Cloud AI platforms

Project What it shows
azure-foundry-policy-agent Azure AI Foundry: Azure OpenAI, hybrid Azure AI Search, Content Safety Prompt Shields, evaluation, managed identity, Bicep
bedrock-aml-triage-agent Amazon Bedrock: AML alert triage with Converse tool use, Guardrails, Knowledge Bases, risk rules, human review, Terraform
vertex-document-intelligence Google Vertex AI Gemini: document classification, schema extraction, validation, confidence routing, EN/AR, Cloud Run
claude-contract-review Anthropic Claude API: contract review with tool use, strict structured output, citations, prompt caching, sign-off audit

Automation and low-code

Project What it shows
n8n-ai-ops-automation n8n: AI ticket triage, PII redaction, human approval for remediation, daily digest, tested governance service
copilot-studio-hr-agent Microsoft Copilot Studio: HR agent for Teams, Entra ID-secured API, Power Platform custom connector, adaptive cards

Private work (demo on request)

Most of my work runs in regulated, air-gapped environments, so the code is private. I'm happy to demo it live and walk through the architecture.

  • Enterprise Agentic AI Factory: private on-prem AI platform covering agents, hybrid RAG, MCP tools with approvals, MLOps with evaluation gates and LoRA fine-tuning, Keycloak, Vault, OpenTelemetry, and two-site HA/DR on OpenShift
  • Secure bilingual AI assistant (EN/AR): fully offline assistant with AD FS/OIDC SSO, secure RAG over approved sources, Planner / Researcher / Writer / Reviewer agents, offline voice, sandboxed tool runner
  • Industry multi-agent portfolio: AML investigation, corporate credit, digital banking, GCC government services, healthcare administration, airline customer care, industrial / physical AI, cyber threat analytics

Tech

Python FastAPI LangGraph LangChain MCP Ollama vLLM Azure OpenAI Azure AI Foundry AWS Bedrock Vertex AI pgvector Qdrant PostgreSQL Kafka Spark Next.js TypeScript Kubernetes OpenShift Docker Terraform Keycloak Vault OpenTelemetry MLflow

Certifications

Azure Solutions Architect Expert · Azure AI Engineer Associate · Azure Security Engineer Associate · Google Professional Cloud Architect · Google Generative AI Leader · OCI Generative AI Professional · OCI Architect Professional · AWS Solutions Architect Associate · TOGAF 10 · SSCP · SAFe 6.0 Agilist

📫 LinkedIn · Open to AI architect and AI engineering roles in the GCC, India and globally

Popular repositories Loading

  1. governed-rag-agent governed-rag-agent Public

    Auditable AI agent: hybrid RAG (BM25 + embeddings + RRF), PII masking, prompt-injection guardrails, human approval for risky tools, tamper-evident audit, eval gates. Zero dependencies, runs offline…

    Python

  2. sarathalapad sarathalapad Public

    Profile

  3. multi-agent-research-crew multi-agent-research-crew Public

    LangGraph multi-agent research assistant: planner, parallel researchers with bounded tool use, writer, self-review loop and human-in-the-loop approval. Runs offline or on local Ollama.

    Python

  4. governed-mcp-server governed-mcp-server Public

    MCP server that gives AI agents governed access to IT operations: risk-tiered tools, policy engine, maker-checker approvals, dry-run, rate limits, emergency lockdown and tamper-evident audit.

    Python

  5. rag-eval-harness rag-eval-harness Public

    Evaluate and compare RAG pipelines and block regressions in CI: hit@k, MRR, nDCG, answer F1, faithfulness, citation validity, abstention, bootstrap CIs. Zero dependencies.

    Python

  6. bedrock-aml-triage-agent bedrock-aml-triage-agent Public

    AML alert triage agent on Amazon Bedrock: Converse API tool use, Guardrails, Knowledge Bases, deterministic risk rules, human review with maker-checker and tamper-evident audit. Terraform included.

    Python