Moraldeep Sachdeo
Engineering Program Manager
About Me
Hi, I'm Moraldeep!
I build AI products that move beyond demos into reliable, real-world systems, with a focus on LLM applications, RAG architectures, and evaluation pipelines.
With 7+ years across Product, Data, Operations, and Engineering functions, I specialize in making AI systems measurable, scalable, and production-ready through strong data strategy and feedback loops. I'm most interested in building AI experiences where trust, performance, and user impact are the core differentiators.
Currently, I am with Microsoft, where I help shape M365 Copilot fine-tuning and design AI evaluation frameworks that improve response quality, reliability, and deployment readiness.
Vibe-Coding
RAG • Routing • Qdrant
A multi-database RAG workflow that classifies questions across product, support, and finance corpora, then falls back to web research when none of the indexed datasets are a good fit.
MCP • Travel AI • Tool Use
A travel planning assistant that uses MCP-connected tools for Airbnb data, Google Maps distances, and live search to generate itinerary-grade recommendations with cost breakdowns and calendar export.
Voice AI • RAG • Streamlit
A voice-first support assistant that crawls product docs, retrieves grounded answers from a vector index, and returns both text and spoken responses in a customizable voice.
Multi-Agent • Research • Workflow
A multi-agent meeting copilot that researches company context, surfaces industry trends, builds a time-boxed agenda, and produces an executive brief before the meeting starts.
AI / ML • Full-Stack Web App
A transparent document verification platform that analyzes uploaded files, surfaces authenticity signals, and explains the reasoning behind each score.
Interactive • Front-End • Simulation
A cinematic, zero-dependency web experience simulating Alex Honnold's free-climb of Taipei 101 with live telemetry, animated atmosphere, and shareable state.
Experience
Aug 2025 – Present
- Owned key workstreams during Microsoft's 1-to-10 growth phase for the M365 Copilot fine-tuning Early Access Program (EAP), leading RLHF and supervised fine-tuning to enhance Copilot's response quality and domain performance
- Designed AI evaluation frameworks and custom model deployment processes, bridging customer needs with core model teams to drive adoption and usage growth
Oct 2023 – Feb 2025
- Launched a 0-to-1 AI-powered conversational suite for enterprise contact centers, including case summarization, smart reply, and auto-compose, driving >25% ticket deflection and $1.2M in annual operational savings.
- Redesigned the RAG stack with metadata enrichment, hybrid BM25 plus embedding retrieval, grounding validation, and search within LLM workflows, reducing irrelevant retrieval by 47%, improving grounding precision by 23%, and halving latency from 1.8s to 900ms.
- Drove LLM fine-tuning for preference alignment and built governance and evaluation frameworks spanning benchmarking, LLM-as-judge scoring, counterfactual and adversarial testing, hallucination stress tests, prompt engineering, and HITL QA validation, cutting fine-tuning cycle time by 35% while improving reliability and BLEU/ROUGE quality.
- Led LLM vendor benchmarking and routing across accuracy, latency, token cost, and throughput; implemented token budgeting, prompt compression, selective grounding, and active-learning labeling pipelines, reducing inference cost by 22% and improving throughput by 3x at enterprise scale.
Jul 2022 – Sep 2023
- Collaborated with Data Scientists and business stakeholders to develop ML predictive models for lead time, cost, and risk, implemented in a SaaS application to optimize Global Procurement vendor selection and negotiations.
- Devised sourcing management strategies and cost-benefit analysis for new product launches, including RFx workflow setup within the Procure-to-Pay system, reducing lead times by 25% and accelerating onboarding of new vendors.
Mar 2021 – Jul 2022
- Reduced supply chain disruptions by building SQL-driven Tableau dashboards that identified root causes of Factory and Operations delays
- Managed creation of a full-stack financial planning platform and led demand forecasting across Product Segments for Global Ops
Aug 2019 – May 2020
- Built Python-driven unsupervised ML models to isolate and identify malicious or fraudulent financial activity
- Designed end-to-end data pipelines converting unstructured text into tokenized, model-ready data
Jan 2019 – Jul 2019
- Optimized video discovery using A/B testing, NLP, and Word2Vec, achieving a 12.5% increase in Click-Through Rate
- Implemented an automated multi-tagging system, enhancing user engagement and sales conversions
Dec 2017 – Dec 2018
- Managed logistics and supply chain operations for 4 Mercedes Benz variants (S, E, C, GLC) at factory level
- Forecasted defects using Regression and Lean Six Sigma, saving $15K+ annually, and presented 20 KPI dashboards to executive committee
Additional Projects
Product Management / Research
A successful attempt to emulate the behavior of a human driver to guess the intention of fellow road users such as pedestrians to cross or not cross the road.
SQL & Lean Six Sigma Methodologies
Research-based project bridging the gap between Value Stream Mapping and Simulation software like Arena to drive continuous improvement in manufacturing organizations and bottleneck identification.
Product Management
Connecting marketplace between HVAC OEMs and buyers. Evangelizing the product vision, developing achievable roadmaps, identifying go-to-market strategies, and forming product alternatives and prototypes.
Education Publications
2019–2020 • Track: Data Science & Product Management
Research: IEEE Publication
Track: Operations Research, Supply Chain Management
Research: IJLSS Publication
Reads
I enjoyed reading these research papers in the last year
Contact
+1 510-646-7721
moraldeepsingh@berkeley.edu
Mountain View, California