Deepak Karkala
Senior ML Engineer | 6 Production ML Systems | GenAI | MLOps | AWS • Python • LLMs | Ex-EPFL • 1 Patent
About Me
Senior ML Engineer with 6+ years of experience in ML/AI, shipping production ML systems that deliver measurable business impact across e-commerce and IoT.
For a mid-sized European e-commerce marketplace, I built 4 production ML systems:
- • 5% conversion uplift via real-time purchase intent scoring
- • 12% higher marketing ROI through CLV prediction
- • 4% search-to-purchase improvement with RAG-powered discovery
- • 60% cost reduction in LLM based review summarization with fine-tuned Mistral-7B
Earlier at eSMART Technologies, I built 2 IoT ML systems:
- • 20% reduction in emergency maintenance callouts via predictive maintenance
- • 10pp increase in solar self-consumption through energy forecasting
Led end-to-end ownership spanning the full MLOps lifecycle: evaluation frameworks (offline metrics, LLM-as-judge, CI/CD regression tests), production debugging (calibration breakdowns, alert fatigue), stakeholder management (legal governance, cross-functional experiments), and operational excellence (latency SLOs, drift monitoring, segment-wise reliability, A/B testing strategy). Delivered cost-bounded infrastructure while maintaining business impact.
Recent GenAI based personal projects include fine-tuning Gemma models: an AI Feynman style Kannada Physics Tutor and an AI Scriptwriter (The Office), both published on Hugging Face with custom reasoning datasets and evaluation harnesses. I also built a prototype Agentic MLOps platform: AI agents that auto-generate complete MLOps infrastructure from requirements to production-ready pipelines with database backed job queues, transparency, guardrails, governance and human-in-the-loop workflows.
For Building Agentic AI Products, I have curated technical strategies on reliability, safety, evals, observability, and governance; business strategies on wedge selection, adoption/distribution loops, pricing, and rollout; plus guides on agent evaluation, agentic RL workflowpolicy optimization, context engineering, and product proposal templates. The Playbooks section distills field-tested patterns and decision frameworks from production systems.
Academic background: MS Communication Systems (EPFL, Grade 5.25/6.0), BE Electronics (RVCE, CGPA 9.23/10.0). Prior work includes research at IISc (Medical Physics publication), signal processing at Signalchip (1 US patent), and ML research at NEC Labs America.
Explore My Work
Past Experiences
Professional ML engineering experience across e-commerce and IoT domains
Senior ML Engineer (Contract) at Mid-sized European E-commerce Marketplace (Client - NDA)
Built 4 production ML systems for a European e-commerce marketplace (75K SKUs, 50K DAU, 2.5K orders/day) driving measurable business impact. Led end-to-end ML lifecycle from problem formulation to deployment and monitoring.
BUSINESS IMPACT:
- 12% higher marketing ROI via CLV prediction
- 5% conversion uplift through purchase intent scoring
- 4% search-to-purchase improvement with RAG discovery
- 60% cost reduction in review summarization
MLOPS OWNERSHIP:
Led full lifecycle: evaluation (LLM-as-judge, CI/CD), production debugging (calibration), stakeholder management (legal, experiments), operational excellence (SLOs, drift monitoring, A/B testing). Cost-bounded infrastructure with business impact.
, [Rumi Schools](https://www.rumischools.org/)](/logos/ventures.png)
Founder at Spiticart, Rumi Schools
Founded two social impact ventures:
- Spiticart - E-commerce platform for traditional Indian handicrafts
- Rumi Schools - Digital-first rural education initiative
 Renens, Switzerland](/logos/esmart.png)
ML Engineer at eSMART Technologies Renens, Switzerland
Owned end-to-end IoT ML systems for smart building operations across 3,000 apartments. Built predictive maintenance and energy forecasting models spanning data quality checks, evaluation, monitoring, and scheduled retraining.
BUSINESS IMPACT:
- Predictive Maintenance: 20% reduction in emergency maintenance callouts
- Energy Forecasting: <10% MAPE energy forecasting accuracy, 10pp increase in solar self-consumption
 Lausanne, Switzerland](/logos/epfl.png)
Master of Science - Communication Systems at EPFL Lausanne, Switzerland
CGPA: 5.25 / 6.0
Master's Thesis: Data Analysis and Anomaly Detection in Buildings Using Sensor Data
Focus: Machine Learning, Applied Data Analysis, Computer Vision, Algorithms
 Princeton, USA](/logos/nec.png)
ML Research Intern at NEC Labs America Princeton, USA
Conducted computer vision research for real-time object detection and scene understanding in traffic surveillance systems.
- Fine-tuned YOLOv2 for tandem motorbike detection in traffic surveillance video
- Built PyTorch relationship detector with scene graph outputs (subject-predicate-object)
](/logos/signalchip.png)
Signal Processing Engineer at Signalchip Innovations
Developed advanced signal processing algorithms for wireless base stations (WCDMA/LTE). Joined as the 5th employee at this semiconductor startup, contributing to core IP.
](/logos/iisc.png)
Junior Research Fellow at Indian Institute of Science (IISc)
Conducted research on diffuse optical tomography at IISc Bangalore.
Published first-author paper in Medical Physics journal: Data-resolution based optimization of the data-collection strategy for near infrared diffuse optical tomography.
](/logos/rvce.png)
Bachelor of Engineering - BE, Electronics and Communication at RV College of Engineering
Focus: Signal Processing, Communications, Algorithms, Mathematics
CGPA: 9.23 / 10.0
Building Agentic AI Products
- Technical Playbook → Production patterns for reliability, safety, evals, observability, and governance
- Business Playbook → How to pick winning agent wedges, design adoption + distribution loops, price safely, and govern rollout
- Evaluation of AI Agents → A Tech-Lead playbook for evaluating agent behavior over time, not just answer quality
- Agentic RL → Optimize the whole workflow policy (tool calls, intermediate decisions) using RFT for measurable lift
- Context Engineering → A systems-first guide to keeping agent context small, relevant, reliable, and auditable
- Product Proposals → 4 domain proposals using a reusable consulting methodology (Product + Tech + Governance)
Product Prototypes
- Agentic MLOps Platform → AI agents that auto-generate complete MLOps infrastructure - from requirements to production-ready pipelines with monitoring, CI/CD, and cost optimization
- AI Scriptwriter for The Office → Fine-tuned open-source LLM that generates authentic sitcom screenplays with character-specific dialogue and comedy beats using RLHF and chain-of-thought reasoning
- Kannada Physics Tutor → AI physics tutor delivering Feynman-style explanations in Kannada language with visual analogies, RAG-powered content, and LLM-as-judge evaluation
Production Playbooks
- MLOps in Production → Comprehensive guide to building, deploying, and maintaining ML systems in production environments with real-world examples and best practices
- AI Agents Playbook → A multi-chapter guide covering the fundamentals and operational practices for deploying autonomous AI agents in production
- GenAI Applications → Practical guide to building production-ready generative AI applications, covering architecture, optimization, and real-world deployment strategies
- Statistics for ML → A practical guide to statistical concepts essential for deploying and monitoring ML systems in production
Agentic Coding
- Spec-Driven Development → Control system designs for turning specs into incremental code changes
- Effective Claude Code → Senior engineer mental models for composable primitives and team workflows
- Claude Agent SDK → Comprehensive guide to building production-ready agentic AI systems using the Claude Agent SDK
- How to Build Coding Agents → Complete playbook covering architecture, training, evals, and observability
Illustrated Guides
- RLHF Illustrated → Interactive web platform transforming complex RLHF concepts into engaging visual learning experiences with 30+ visualizations
- Stats for MLOps Illustrated → Interactive platform teaching production ML statistics through narrative storytelling, live visualizations, and real-world case studies
Work With Me
I bring hands-on experience delivering production MLOps and GenAI systems at moderate scale—with minimal infrastructure footprint and cost-effective architectures. I'm excited to collaborate on building next-generation Agentic AI systems. Whether you need expertise in MLOps, GenAI, or Agentic AI—let's connect.