- Building and deploying an agentic AI assistant for Amazon sellers โ a conversational system that queries Seller Central data (inventory, orders, revenue, shipping) and delivers business insights in natural language with inline visualizations
- Designed the full RAG pipeline: SP-API data ingestion โ ETL โ PostgreSQL (structured) + ChromaDB (vector embeddings) โ LangGraph agent with SQL and retrieval tools โ Claude/GPT response generation
- Implemented safety and guardrails layer โ query classification, harmful content detection, policy-bypass prevention, and hallucination reduction achieving 99% response accuracy
- Leading RLHF evaluation workflows: rating helpfulness, factual accuracy, and data correctness across 1000+ query-response pairs to continuously improve model quality
- Benchmarking AI models end-to-end โ defining test plans, evaluation metrics, and acceptance criteria across accuracy, robustness, bias, and reliability
๐งฉ Tailored for roles: skills and projects are ordered for this role, nothing is hidden. Standard view

SUDISH KUMAR
ML Engineer | Deep Learning (PyTorch) | Generative AI | AWS SageMaker
I build AI systems that actually ship to production โ not just notebooks that collect dust. Currently at Amazon deploying agentic AI and benchmarking LLMs. Two AWS certifications, 99%+ accuracy across 10+ initiatives, and tools that save my team 150+ hours daily. Looking for my next challenge.
Impact
By the Numbers
Technical Arsenal
What I Work With
Machine Learning & Deep Learning
Generative AI & Agentic Systems
AWS & MLOps (Production)
Python (Production-Grade)
SQL & Data Engineering
Analytics & Visualization
Tools & Production Infrastructure
Web Development (Side)
Where I've Worked
Work Experience
- Built automated data pipelines using Python (Pandas, NumPy) and SQL to prepare, validate, and annotate large-scale datasets across text, code (QAL), and computer vision (image/video) โ maintaining 99%+ accuracy across all deliverables
- Engineered high-precision training datasets for Agentic AI models across Finance, Mortgage, Health, and Legal domains โ 10+ initiatives including Red-Teaming, mathematical evaluation, and multi-turn conversation flows
- Developed "SageMaker Smart Stats" from scratch โ a JavaScript-based productivity analytics tool that automated metric tracking, surfaced workflow bottlenecks, and saved ~45 minutes per associate daily across a 200-person team (15-20% output increase)
- Authored technical SOPs and annotation guidelines for complex ML tasks โ reduced guideline-related errors by 15% and significantly cut rework cycles, becoming the go-to documentation resource for the team
- Built custom Tableau and Python dashboards for data quality monitoring โ flagging annotation drift, inter-annotator agreement scores, and SLA adherence in real-time
- Built Python-based automation scripts for data validation and quality scoring โ improved AI model training accuracy from 94% to 99% while reducing manual rework by 30%
- Developed performance monitoring dashboards using Python (Pandas, Matplotlib) and Excel VBA โ tracking per-associate throughput, error rates, and SLA adherence in real-time
- Maintained 99.5% dataset accuracy across all deliverables with zero SLA misses โ identified systematic labeling errors through statistical analysis and implemented corrective feedback loops
- Taught Physics, Mathematics, Python, and SQL to 500+ students. Average scores above 91%. Explaining complex technical concepts clearly โ that's a transferable skill that shows up in my documentation, SOPs, and team communication
Featured Work
Projects
From model training to production APIs โ here's what I've shipped
Real-Time Stock Anomaly Detector
Pulls live market data via API, applies Z-score and rolling-window anomaly detection, flags unusual price movements in real-time. Clean OOP structure with type hints, full test coverage. Visualizes anomalies with interactive Plotly charts.
E-Commerce SQL Analytics Engine
Advanced SQL analytics on real e-commerce data: cohort retention analysis, RFM customer segmentation, funnel conversion tracking, revenue forecasting with window functions. Includes EXPLAIN plan optimization and indexing strategy.
GenAI Multimodal Chatbot โ AWS Bedrock
Full-stack generative AI chatbot handling text, images, and documents. Claude via Bedrock with streaming, conversation memory, and enterprise guardrails. Serves 500+ concurrent users.
RAG Document QA โ AWS Bedrock
Production RAG pipeline that cites its sources. Semantic chunking, FAISS vector store, Bedrock embeddings. Handles enterprise document loads, not a toy demo.
End-to-End ML Pipeline โ SageMaker
Full MLOps for fraud detection: automated feature engineering, hyperparameter tuning, A/B testing, real-time inference endpoint. 96.3% precision on prod data.
LLM Fine-tuning โ PyTorch (LoRA)
Fine-tuned Llama-2 with LoRA/QLoRA for financial QA. 4-bit quantization, runs on consumer GPUs. 23% accuracy improvement over base model.
Sentiment Analysis API โ Transformers
Real-time sentiment using fine-tuned DistilBERT. Dockerized, deployed on Lambda. Sub-100ms latency, 1000+ req/min. A model is useless if you can't serve it.
Image Classification โ CNN + ResNet50
Multi-class medical image classification with transfer learning. Augmentation pipeline, 94.7% test accuracy. REST API on EC2 for real-time predictions.
SageMaker Smart Stats
Built at AWS because the team needed it. Real-time productivity metrics, surfaces bottlenecks, visualizes trends. 200-person team uses it daily. Saves ~45 min per person.
Enterprise Fiber Management System
Full-stack telecom network ops platform (BSNL-style). Fiber routes, splice points, customer connections. Complex topology mapping and real-time status tracking.
Freelance Web Design Portfolio
Business websites for local clients โ mehndi artist, hospital. Responsive, mobile-first, SEO optimized. Sometimes you just want to make things look good.
Robotic Car โ Raspberry Pi
Pi-based robot with 4-motor drive. Python control with sub-50ms latency. 30% component reduction through circuit optimization. Software meets hardware.
Nothing in this category yet.
Shipped & Running
Live Projects
Live from GitHub
Shipping in Public
Fetched straight from the GitHub API as you open this page โ the most recently pushed public repositories, no CMS in between.
Everything else lives at github.com/Sudish007 โ
The Pitch
Why Hire Me?
What I bring to your team that most candidates don't.
I Ship, Not Just Experiment
Most ML engineers stop at the notebook. I take models from research to production โ with monitoring, testing, and real users.
99%+ Accuracy, Every Time
10+ AI initiatives, near-perfect data quality. Your model is only as good as your data. I don't cut corners.
Tools That Scale to 200+
SageMaker Smart Stats saves 45 min/day per person across a 200-person team. I solve problems for everyone, not just myself.
Recognized Performer
Prime Player Award twice in 3 months. Multiple leadership shout-outs. When something needs to be done right and fast, I'm the one they call.
Full ML Lifecycle
Data engineering โ training โ deployment โ monitoring. No hand-offs needed. I own the entire pipeline end to end.
Clear Communicator
7 years explaining physics to students. SOPs that 200 people follow. I can explain your model to the CEO and the intern.
What colleagues say
Recommendations
๐ฅ Want Me on Your Team?
I'm actively looking for my next ML/AI engineering challenge. If your team builds intelligent systems, ships models to production, or needs someone who combines technical depth with clear communication โ let's talk.
Credentials
Certifications
Recognition
Awards
Prime Player Award โ Feb & Apr 2025
Highest productivity and reliability across the entire team. Twice in 3 months.
Leadership Recognition
Multiple shout-outs for Agentic AI SOP development, multilingual dataset creation, and going above and beyond.
Let's Connect
Get In Touch
Looking for AI/ML roles, data science positions, or freelance consulting. Pick your preferred way.
๐ผ Also Available for Freelance AI/ML Consulting
Need a custom ML model, data pipeline, or AI tool? I take projects from idea to production. Let's talk scope and timelines.
Currently Exploring