AI engineer · gen AI · agentic systems
Vikash Kumar
I build AI systems that survive contact with production — not demos.
I build multi-agent workflows, agentic RAG pipelines and LLM features for security operations at Armor — and full-stack products either side of that. Python and TypeScript, with a bias toward things that ship.
What I actually work on
The model is rarely the hard part. These are the four things that decide whether an AI feature is trustworthy enough to put in front of a user.
Agentic systems
Multi-agent workflows with planner, worker and reviewer roles — plus the failure recovery, timeouts and liveness checks that keep long-running jobs from silently stalling.
RAG that retrieves
Ingestion through chunking, embeddings, vector search and reranking. Retrieval quality is the product; the model is the easy half.
Tools & MCP
Exposing real systems to agents as typed, well-described tools, so an agent can act on production without being trusted blindly.
Evals & reliability
Regression gates on agent behaviour, observability into every step, and cost control — the unglamorous work that makes AI output trustworthy.
Featured projects
All projects →
Agentic AI — Multi-Agent & MCP Projects
Jun 2025 — presentA set of agent systems built across the four stacks that matter in production — OpenAI Agents SDK, LangGraph, CrewAI and MCP — each one built twice to learn where the abstractions help.
PythonOpenAI Agents SDKLangGraphCrewAI+4
AI Attack-Story Timeline
Mar 2025 — presentAn AI-assisted feature for a security platform that correlates a flood of detections into one analyst-readable timeline of what an attacker did, and in what order.
PythonLLM orchestrationAgentic RAGVector search+2
SharedReads
Jul 2026 — presentA book-sharing platform designed, built and deployed solo — React and TypeScript front end, Node API, PostgreSQL, self-hosted behind nginx and a Cloudflare Tunnel.
ReactTypeScriptNode.jsExpress+4
Experience
Full detail →Software Developer II · Armor Defense
Mar 2025 — presentPune, India
Building an AI/agentic platform for security operations, plus the full-stack product surface around it.
Software Engineer · Volkswagen Group Technology Solutions India
Oct 2023 — Mar 2025Pune, India
Full-stack engineering on Volkswagen Group's internal engineering and delivery platforms.
UI Developer · Adeption
Jul 2022 — Sep 2023Pune, India
Front-end and full-stack development on a SaaS leadership-development platform.
Software Engineer · BigBinary
Oct 2020 — Jun 2022Remote
Full-stack product engineering for US-based clients in a fully remote, high-autonomy consultancy.
Tools of the trade
- AI & LLM
- PythonLangChainLangGraphCrewAIOpenAI Agents SDKMCPPyTorchFine-tuning (QLoRA)NLPPrompt engineeringMulti-agent orchestration
- RAG & retrieval
- Chunking strategiesEmbeddingsVector searchpgvectorPineconeQdrantHybrid searchReranking
- LLMOps
- Evals & regression gatesLangSmithLangfuseRAGASGuardrailsTracing & observabilityToken & cost control
- Backend
- PythonFastAPINode.jsTypeScriptPostgreSQLRESTGraphQLWebSocketsDistributed systemsData pipelines
- Cloud & platform
- AWS (Bedrock)Azure OpenAIVertex AIOracle Cloud (OCI)DockerKubernetesCI/CDnginx
- Frontend
- ReactNext.jsTypeScriptTailwind CSSFlutter
Get in touch
Open to AI Engineer, Gen AI Engineer and Forward Deployed Engineer roles. FDE work is where I'm most at home — sitting with a customer's real problem, building the thing, and staying on it until it works in their environment.