AI Engineer with 2.5+ years building end-to-end production AI systems — from GPU-accelerated LLM inference and RAG pipelines to GAN-based hallucination mitigation and multi-agent automation.

I work across the full hardware-to-software AI stack: selected globally for Google Summer of Code 2026 at Intel/OpenVINO to build a 5-agent privacy-preserving desktop automation system, and contributed 12 verified merged PRs into Intel's production inference toolchain, the Linux Foundation's canonical RISC-V ISA specification, and Harvard/MIT's HNN-Core.

May 2026 – Sep 2026  ·  Google-Sponsored · Remote

Selected globally to build a privacy-preserving GUI Desktop Automation Agent — fully local, zero cloud dependency.

5-Agent Pipeline (A2A Protocol + MCP Server)
─────────────────────────────────────────────────────────────────
Router → Planning → UI Grounding → Action Execution → Reflection
         DeepSeek-R1    Phi-3.5-Vision      Native OS         Self-Correction
         Qwen-7B-INT4   INT4 via OVMS       via MCP           Loop

• Sub-600ms step latency · INT4 quantization · KV caching · prefix caching · WEIGHTLESS optimization
• Mentored directly by Intel engineers Ethan Yang and Zhuo Wu
• Entire pipeline runs on-device — enterprise-grade privacy by design

Apr 2026 – Present  ·  Walnut Creek, CA · Remote

• Production LangChain RAG pipelines with FAISS for enterprise knowledge management — measurable gains in LLM accuracy and hallucination reduction
• Full data engineering stack: ingestion → semantic chunking → embedding → FAISS indexing → retrieval
• Agentic AI workflows for autonomous multi-step task execution
• Owns model monitoring loops — retrieval quality, relevance, and latency metrics

Jan 2026 – Apr 2026  ·  Remote

• Built Verimate — AI platform automating UVM verification plan and testbench generation for RTL/chip designs (semiconductor sector)
• LangChain + FAISS + Gemini API + GPU-accelerated local LLMs
• Delivered measurable reduction in hardware verification engineering cycle time

Dec 2025 – Mar 2026  ·  Santa Clara, CA · Remote

• Transpose-aware LoRA Correction — correct handling of MatMul layers with transposed activations in INT4/INT8 quantized LLMs
• Migrated LLM compression examples to OpenVINO stateful inference flow — production-grade, no manual KV-cache handling
• 3 merged PRs: NNCF #3814 · NNCF #3864

Mar 2025 – Jun 2025  ·  Remote · Mentored by Qualcomm, Ventana, Synopsys

• 7 merged PRs to the canonical machine-readable RISC-V ISA specification
• Implemented ISA extensions: Zilsd, Zclsd, Zcmop compressed MOP instructions
• Ruby tooling, CI/CD automation with GitHub Actions, schema and IDL fixes

PRs: #521 · #530 · #542 · #577 · #617 · #654 · #923

Jul 2023 – Dec 2025  ·  Karachi, Pakistan

• Led AI4Org: multi-discriminator GAN + REINFORCE RL hallucination mitigation — fine-tuned TinyLlama / GPT-2 / LLaMA2 / Mistral 7B with PEFT/LoRA on dual AMD RX 7900 XTX
• Led ArcheV: first LLM benchmark suite for RISC-V RV32I assembly code — functional correctness, syntactic validity, ISA edge-case evaluation
• Built Vermithor: 5-stage pipelined RV32I processor in Chisel (Scala) — hazard detection, forwarding, Verilator verification

AI / LLM

GPU & ML

Cloud & MLOps

Languages

BSc Software Engineering — UIT University, Karachi  ·  Feb 2022 – Feb 2026

IBM AI Engineer Professional Certificate (Coursera / IBM, 2025)

Generative AI Applications with RAG and LangChain  ·  Fundamentals of AI Agents Using RAG and LangChain  ·  Generative AI Advanced Fine-Tuning for LLMs  ·  Generative AI Engineering and Fine-Tuning Transformers  ·  Deep Learning with PyTorch  ·  AI Capstone Project with Deep Learning

"The gap between a model and a system is where most people give up. That's where I live."