Message ready!
ML Engineer · GenAI · AI Systems

Ashish Varshney

Building production-grade ML systems & LLM pipelines that scale, explain themselves, and keep getting better.

Ashish Varshney
Open to Roles
3+
Years Experience
60%
Debug Time Reduced
3+
Production Systems
01 /

About

Senior AI Engineer with 3+ years of experience building production-grade Python backend systems and end-to-end ML pipelines in enterprise environments. I thrive at the intersection of GenAI innovation and robust system engineering.

Experienced in Flask-based API engineering, LLM-powered log intelligence, MLOps automation, model deployment, and observability platforms. Proven track record delivering scalable ML solutions for anomaly detection, failure prediction, and automated remediation using state-of-the-art GenAI and deep learning frameworks.

Strong ML/DL/GenAI foundation from IIT Roorkee, with deep hands-on expertise across the full ML lifecycle — from data engineering and model training to deployment, monitoring, and CI/CD automation. Passionate about systems that don't just work — they scale, explain themselves, and keep getting better.

⚡ LLM Pipelines 🧠 GenAI 📦 MLOps 🔭 Observability 🏗️ System Design ◆ DSA
02 /

Experience

Sep 2024 – Jan 2026
HCLSoftware
Technical Lead — Backend & AI Systems
BigFix Platform
  • Architected a production-grade LLM-powered log intelligence pipeline using Python, regex NLP normalization, OpenAI/HF embeddings, and FAISS vector search to cluster recurring log patterns and detect failures. ↓70% noise
  • Built an advanced RAG-based Root Cause Analysis engine using Retrieval-Augmented Generation, prompt-chaining, and context optimization. ↓60% debug time
  • Developed fully automated CI/CD pipelines using Python scripts, Jenkins, and n8n orchestration covering environment setup, cross-OS test execution, versioning, and release automation. ↑40-50% faster releases
  • Engineered large-scale configuration and remediation engines for RedHat OS, databases, and middleware stacks, reducing manual remediation workload through automated fixlet generation. ↓60% manual work
Jul 2023 – Aug 2024
HCLSoftware
Technical Lead / ML Engineer
Telcom Observability Platform
  • Designed ML-based anomaly detection and failure prediction systems using ARIMA, Isolation Forest, and XGBoost for multivariate telemetry streams from VM servers and data-center infrastructure. ↓40-60% MTTD
  • Led end-to-end migration of high-throughput telemetry pipelines from GCP Pub/Sub Lite to Kafka and Apache NiFi, improving ingestion scalability for thousands of nodes.
  • Built low-latency Flask APIs to collect telemetry from OpenNMS collectors across VM servers and data-center nodes, enabling real-time KPI computation and proactive incident detection.
  • Delivered rapid cloud-to-on-prem migration of a production data ingestion platform using AI-assisted GitHub Copilot prompts, significantly reducing cloud costs and enabling full production deployment within days.
  • Boosted engineering productivity by implementing AI-generated API unit testing workflows, reducing manual testing effort, accelerating release cycles, and minimizing production defects.
03 /

Skills

⚡ Generative AI
LLMs (GPT-4o, Claude, Gemini)RAGPrompt EngineeringLangChainLangGraphLangsmithLlamaIndexAI AgentsVector DBsFAISS / ChromaHuggingFaceFine-tuning (LoRA/QLoRA)EmbeddingsPrompt Chaining
Languages
PythonC++CSQLBash / Shell
ML & AI
Anomaly DetectionXGBoostRandom ForestARIMAIsolation ForestFeature EngineeringScikit-learnNLP / Text MLTime-Series Forecasting
Deep Learning
TransformersCNNsRNNs / LSTMsAttention MechanismsTensorFlowPyTorchKerasTransfer LearningGrad-CAMEfficientNet / ResNet
◆ DSA & Problem Solving
Dynamic ProgrammingGraph AlgorithmsTrees & HeapsSorting & SearchingSliding WindowRecursion & BacktrackingHashingGreedy Algorithms
MLOps & Deployment
MLflowBentoMLModel ServingModel MonitoringJenkinsn8nDockerCI/CD Pipelines
Backend & Systems
FlaskFastAPIREST APIsFAISSKafkaApache NiFiGCP Pub/SubMySQLRedis
Data Science
PandasNumPyMatplotlibSeabornEDAStatistical Modeling
04 /

End-to-End Architecture

From raw data ingestion to production inference — I design and own the full ML system lifecycle. Below are the key architectural patterns and pipelines I've built and shipped at enterprise scale.

⚡ LLM-Powered RAG Architecture — Log Intelligence & Root Cause Analysis
📥Raw Logs
🔧NLP Normalize
🧮Embeddings
🗂️FAISS Index
🔍Semantic Search
🤖LLM + Prompt Chain
📊RCA Output
🔬 ML Anomaly Detection Pipeline — Telcom Observability Platform
📡Telemetry Stream
⚙️Kafka / NiFi
🧹Feature Eng.
🌲Isolation Forest
📈ARIMA / XGBoost
🚨Alert Engine
🎛️Flask API
🧠
GenAI System · Production
RAG-Based Intelligent Log Analysis

Production LLM pipeline combining FAISS vector search, OpenAI/HuggingFace embeddings, and prompt-chaining for automated root cause analysis on enterprise log data. Reduced mean debug time by 60% and eliminated alert fatigue across teams.

LangChainFAISSOpenAI APIHuggingFaceFlaskPython
🔭
MLOps · End-to-End
Full ML Lifecycle — Telcom Observability

Complete ML system from data ingestion (Kafka, NiFi) → feature engineering → model training → MLflow tracking → BentoML serving → Flask API → monitoring dashboards. Live on thousands of infrastructure nodes.

KafkaApache NiFiMLflowBentoMLXGBoostFlask
🏗️
Deep Learning · Computer Vision
Explainable CNN Pipeline for Medical Imaging

End-to-end X-ray analysis system: data preprocessing → EfficientNet/ResNet with custom CNN layers → Grad-CAM visualization for clinical interpretability → production inference API.

PyTorchEfficientNetGrad-CAMOpenCVFastAPIDocker
⚙️
CI/CD · Automation · GenAI-Augmented
AI-Augmented DevOps & Release Automation

Automated CI/CD architecture using Jenkins + n8n + Python covering: environment provisioning, cross-OS test execution, AI-generated unit test suites via GitHub Copilot, versioning, artifact publishing, and deployment gates. Cut release cycles by 40–50%.

Jenkinsn8nGitHub CopilotPythonDockerShell
05 /

Writing

I write about AI, machine learning systems, and the craft of building production-grade software. Published on Substack — thoughts on GenAI, MLOps, and the future of intelligent systems.

SubstackNote · AI Systems
Building Smarter AI — Note #1

Thoughts on designing intelligent systems that go beyond the hype — practical insights into LLM pipelines, RAG architectures, and what it actually takes to ship GenAI in production.

Read on Substack →
SubstackNote · MLOps
Building Smarter AI — Note #2

Exploring the full ML lifecycle: from telemetry ingestion and anomaly detection to production deployment. Real lessons from building observability platforms at enterprise scale.

Read on Substack →
06 /

Projects

IIT Roorkee · Deep Learning
Explainable CNNs for COVID-19 Diagnosis

End-to-end X-ray COVID-19 detection pipeline using EfficientNet/ResNet with custom CNN layers. Applied Grad-CAM to visually highlight infected lung regions, improving model interpretability for clinical validation workflows.

IIT Roorkee · Predictive ML
E-commerce Price Prediction System

Complete price prediction ML pipeline with advanced feature engineering across brand, CPU, RAM, and GPU attributes. Achieved 15% accuracy improvement through systematic hyperparameter tuning using XGBoost and Random Forest optimization.

HCLSoftware · Cloud Migration
Cloud-to-On-Prem Data Ingestion Migration

Rapid migration of a production data ingestion platform from GCP Pub/Sub to Kafka & Apache NiFi using AI-assisted GitHub Copilot prompts. Significantly reduced cloud infrastructure costs and achieved full production deployment within days.

HCLSoftware · GenAI · QA Automation
AI-Generated API Unit Testing Workflows

Implemented AI-powered API unit testing pipelines that automated test generation and execution using LLM prompt engineering and GitHub Copilot. Reduced manual testing effort substantially, accelerated release cycles, and minimized production defects.

07 /

Education

M.Tech in Artificial Intelligence
Indian Institute of Technology, Roorkee
B.Tech in Computer Science
ZHCET, Aligarh Muslim University
08 /

Achievements

🏆
Swachh Bharat Internship — 2nd PrizeNational digital awareness campaign — MHRD. Awarded for blogs, presentations, videos, and online content.
🎓
IIT Roorkee — M.Tech AISecured admission into the prestigious AI program at IIT Roorkee through competitive entrance examination.
Production Impact — 60% Debug Time ReductionDelivered measurable enterprise-scale impact through GenAI-powered RCA engine shipped in production at HCLSoftware.
09 /

Get in Touch

Open to exciting new opportunities in ML Engineering, GenAI Systems, and Backend roles. Whether you have a project in mind, want to collaborate, or just want to talk AI — I'm all ears. Let's build something incredible together.

Send a Message
// I'll respond within 24 hours
Please enter your first name.
Please enter a valid email address.
Please enter a subject.
Please enter a message (at least 20 characters).