I build intelligent systems. The rest is figuring out what comes next.

Shikhar. Working at the intersection of AI, automation, and the systems that will define the next decade.

reasoning-engine.sh

Tech Stack

n8nVapiQdrantDockerRAG PipelinesLLM OrchestrationWebhook ArchitectureVector EmbeddingsPythonJavaScript

What I Work On

Live

Enterprise RAG with RBAC & Hybrid Search

Production-grade retrieval system combining BM25 keyword search, dense vector search, and cross-encoder reranking — with role-based access control that filters documents before the LLM sees them. Built to production standards — RBAC, hybrid search, and full RAGAS evaluation.

QdrantBM25Cross-EncoderPython
Live Demo →

Self-Healing Multi-Agent DevOps Pipeline

A multi-agent system that monitors a codebase for errors, decomposes the fault, proposes a fix, tests it in an isolated Docker environment, runs security scans, and only opens a Pull Request when everything passes.

LangGraphSentrySemgrepDockerGitHub Actions

Lead-to-Call Automation

Full-stack sales pipeline built on n8n — captures leads via webhook, scores them with Gemini AI, routes by priority, triggers personalized email sequences, updates HubSpot, and escalates high-value leads to an AI voice agent that calls and qualifies them in real time.

n8nGeminiVapiHubSpotWebhooks
Currently Building

Edge Voice Assistant

Local-inference voice agent running quantized LLMs on VPS via Ollama with WebSocket streaming — targeting sub-200ms first token latency with full STT → LLM → TTS pipeline.

OllamaWebSocketsWhisperKokoro TTSPython

Thinking

The Panic Is Aimed at the Wrong People

Everyone is panicking about AI taking jobs broadly. The real displacement is narrower — any role that runs on pattern matching and low-context decisions is already being automated. The panic is misdirected. The people who lose jobs aren't the ones adapting slowest, they're the ones whose entire role was never really cognitive work to begin with.

LLMs Are Not the Destination

Next-token prediction scales well but it's not reasoning — it's sophisticated pattern completion. LLMs don't model causality, they model correlation at massive scale. Something fundamentally different needs to happen architecturally before machines can actually think. The real breakthrough hasn't been invented yet.

The Jobs That Survive Are the Ones AI Can't Simulate

The roles that outlast automation aren't just the high-cognition ones — they're the ones where the training data doesn't exist and the hardware isn't there yet. Blue collar trades, physical dexterity, real-world context. And on the other end, the people bridging AI to existing human infrastructure. The middle gets hollowed out. The edges survive.

Let's Talk

If you're building something interesting, thinking about where AI is going, or want to work together — reach out.

Send a message