Ashish Anton Abraham

AI Engineer specializing in LLMs, Agentic AI, and production-grade ML systems, with hands-on experience building and deploying scalable AI solutions. Currently working on edge-deployed real-time LLM pipelines for in-vehicle experiences, along with agentic systems.
Strong background in machine learning research, with published work in NLP.

Experience

AI Engineer-1

Aug 2024 - Present
Visteon Corporation, Trivandrum
  • Developed embedded large language model (LLM) pipelines for in-car intelligence, achieving sub-100ms local inference on Qualcomm and Nvidia chipsets through KV-cache and kernel-level optimizations. Combined Apple Fast-ViT encoder with Qwen2.5 decoder to achieve around 70ms inference on device.
  • Designed agentic pipelines that reduce testing and triaging time by 70% and improve employee productivity by 50% through automation and workflow optimization.

SDE Intern

Jan 2024 - Jun 2024
Visteon Corporation, Trivandrum
  • Engineered interfaces in Android Open Source Project (AOSP) for updater services and over-the-air updates, contributing code deployed in over 300,000 vehicles.
  • Supported large-scale system deployment and codebase management for automotive software solutions.

ML Intern

Jun 2023 - Aug 2023
Sensegrass AI, USA
  • Fine-tuned GPT-3 large language models to build a fully automated hiring tool, reducing manual HR workload by 70%.

ML Engineer - Contractor

Jun 2022 - Dec 2022
Snap Inc., USA
  • Developed machine learning-powered augmented reality lenses, optimizing GAN models for mobile deployment by reducing memory usage by 40%.
  • Focused on model compression and real-time performance for AR applications.

SDE Intern

May 2021 - May 2022
Phorina, USA
  • Optimized network modules in Rust using MQTT-SN protocol, reducing code size by 40% and improving memory usage and speed by 25% for scalable IoT solutions.

Research

2026

Dual Encoder Fusion in a Hierarchical Transformer with Lexical Syntactic and Semantic Attention for AI Paraphrase Detection, ACL (Accepted, to appear)

2025

An Adaptive Learning System for Medical Education using Knowledge Graph and Retrieval-Augmented Generation, IEEE Xplore (Accepted, to appear)

Projects

SketchGAN

Implemented SketchGAN, a PyTorch-based cascade encoder-decoder network for joint sketch completion and recognition, completing missing portions of input drawings iteratively. Developed and trained generative adversarial networks for image completion tasks.

QueReyDB

Developed a Retrieval Augmented Generation system integrating vector search to translate natural language into SQL queries for PostgreSQL, leveraging large language models to improve query accuracy and efficiency by 40%. Designed and implemented an end-to-end NLP pipeline for translating user queries into SQL.

Contact

Feel free to reach out via email for opportunities or collaborations.

> whoami