Hi, I'm Abhishek.

Software developer · AI Pragmatist · Systems Architect · Artist

Bridging deep tech infrastructure and creative exploration, I engineer core systems to modernize complex, multi language codebases. From designing deterministic AI to real time MR fusion, I architect resilient foundations that resolve structural challenges and unlock product stability.

As technology evolves, I am shifting from solving problems to anticipating them through first principles design. I envision a future where intelligent systems act as invisible pillars of progress, scaling beyond automation to solve global challenges. My mission is to bridge human intent and machine intelligence, crafting infrastructure that delivers lasting value to the digital landscape.

Experience

Present July 2019

Lead Software Engineer

July 2019 - Present

Onix

Lead a team of 6+ engineers of L3 - L4 level to drive AI-native development, utilizing Modern AI tools to accelerate the delivery of deterministic processing engines and reduce QA cycles by ~40%.
Architected large-scale data systems to extract execution lineage and dependency graphs for Shell, SQL, JavaScript, Python, and Scala, utilizing Neo4j to drive modernization strategies.
Achieved 100% analysis coverage (up from 60%) by designing runtime engines with control-flow tracking and runtime interception using GraalVM and Truffle instrumentation.
Improved lineage reliability by ~35% by building a confidence-based trust scoring system leveraging RAG-based context enrichment.
Engineered a unified logic interpretation layer using the Apache Calcite framework and a custom canonical model to handle diverse input representations across multi-language codebases.
Developed a 2-step Scala lineage pipeline using a hybrid LLM approach: automated parsing/extraction to create a DTO model, followed by Semantic Search on a Vector DB to inject dependency source code into detailed prompts for commercial llm processing.
Delivered high-impact PoCs and critical fixes, including Databricks, Airflow metadata extraction, ANTLR-based TPT parsers, and resolving production blockers during legacy Shell-to-Java upgrades.
Spearheaded core infrastructure improvements, including containerization, CI/CD optimization, and a plugin-based architecture, reducing deployment time by ~30%.
Drove AI enablement initiatives by integrating Model Context Protocol (MCP) with internal metadata glossaries, providing LLMs with context aware understanding for explainable lineage and product insights.

Java Kotlin Neo4j Shell Scripting GraalVM RAG MCP Docker

Projects

Mixed Reality Exploration

Dec 2022

I wanted to explore Mixed Reality (MR) at a time when the field was still largely experimental for mobile. The goal was to break the 2D screen barrier using only a standard smartphone, shifting the interface from "touch" to on-device AI and gesture-driven interaction.

The Idea: The project focused on shifting the user interface from traditional "touch" to gesture-driven interaction. By treating the smartphone camera as a spatial sensor, I developed a way to overlay and manipulate digital objects in the physical world, prioritizing a "physics-first" feel where digital elements responded naturally to human intent.

The Tech: I used Mediapipe for hand-landmark detection and OpenCV to process the live camera feed. By building the pipeline to run entirely on-device, I eliminated cloud latency, ensuring the 3D objects reacted instantly to movements. I solved the "floating" problem by calculating 3D coordinates from 2D camera points in real-time—balancing complex math with the limited CPU power of a mobile device to keep the frame rate smooth.

Android Mediapipe OpenCV

Logistics Platform – Local Delivery App

July 2017

I developed a logistics platform to coordinate real-time commerce between local vendors, delivery partners, and customers. The goal was to build a system to manage the entire order lifecycle from initial placement to doorstep delivery within a single digital ecosystem. Built in 2017, this was an early exploration into the "quick-commerce" models that have since become industry standards.

The Idea: I envisioned a platform where consumers could order goods directly from their choice of local shops and receive home delivery with features like real-time tracking. The core concept focused on centralizing the logistics chain, allowing a seamless flow of information from the moment a customer placed an order to the second it reached their door.

The Tech: I built a dedicated Store Owner’s Dashboard for order management and a Consumer App for browsing and tracking. I used Node.js for the backend logic and Firebase to handle real-time data synchronization. By integrating the Google Maps API, I enabled live transit telemetry, ensuring the consumer app received millisecond-accurate location updates while the dashboard maintained an overview of all active deliveries.

Android Node.js Express MySQL Google Firebase

NotifyD – Lightweight System Monitor

April 2016

I built NotifyD to solve the problem of manual server oversight. It is a daemon-based monitoring tool designed to track system health in the background and provide instant visibility into resource spikes or failures without the complexity of enterprise-grade suites.

The Problem: As a student, I needed a way to monitor my servers without paying for expensive third-party tools. I faced a specific constraint: I didn't want to keep my mobile data on at all times just to receive alerts. I needed a way to get instant SMS notifications directly to my phone so I could stay informed about system health even when I was offline.

My Solution: I engineered a monitoring daemon that tracked system metrics and triggered alerts via Firebase Cloud Messaging. I built a custom Android App to receive these push notifications; if the app detected that the device was active, it displayed an alert. I also developed an Admin App that acted as a gateway, automatically converting those Firebase signals into SMS messages to ensure I received the notification regardless of my internet connectivity.

Android Firebase Node.js shell scripting

Writings

Dec 2025

Designing Intent-Based Search for Unstructured Media

Social platforms are built for discovery, not retrieval. I explore a technical framework using AI enrichment and hybrid search—to turn "lost" saved content into a searchable personal knowledge.