Hi, I'm

Manish Joshi!

I'm a Senior Data Scientist at

Presight AI.

7+ years · Microsoft → TikTok → Presight AI · Billion-scale ML & AI systems

Connect to explore solutions and customized products
—designed for you !

Open to work

Built systems at

Key Expertise

Led design & delivery of agentic AI systems & multi-agent architectures for enterprise clients

Deployed on-prem LLMs (MiniMax-M2.5, SLMs) with KV cache optimization for maximum throughput

Improved TikTok Recommendation & Notification algorithms at billion-user scale

Shipped end-to-end systems at Bing scale across multiple markets

Skills

AI Agents

LLM Deployment

RAG

Recommender Systems

Machine Learning

Deep Learning

vLLM & llama.cpp

Python

LangGraph

LLM Function Calling

NLP

KV Cache Optimization

PyTorch

MLOps

Awards & Achievement

Achievement Beyond Excellence Award

TikTok • 2021 • Engineering Excellence

Microsoft Patent Award

USPTO Patent • 2020 • "Visual Parsing for Annotation Extraction"

Excellence in Innovation Award

Microsoft FY2020-Q2 • "Program Synthesis for Annotations"

GATE All India Rank 28

2017 • Computer Science

Professional Journey

Senior Data Scientist

Presight AI

Feb 2024 - Present · 2 yrs 1 mo

Abu Dhabi Emirate, UAE · On-site

• Led multiple agentic AI systems for enterprise — research, analytics & on-prem Copilot alternatives

• Built Code-Agent with sandboxed execution, chart generation & predictive modeling

• Architected agentic search as embedding-free RAG alternative

• Deployed on-prem LLMs (MiniMax-M2.5, SLMs) via vLLM & llama.cpp with KV cache tuning for max throughput

• Designed LangGraph multi-agent architectures with LLM-as-Judge evaluation

Agentic AIOn-Prem LLM DeploymentvLLM & llama.cppKV Cache OptimizationLangGraphLLM-as-Judge

Machine Learning Engineer

ProShort.ai

Sep 2022 - Jul 2023 · 11 mos

Bengaluru, Karnataka, India · Hybrid

• Implemented recommendation algorithms from scratch with real-time updation across multiple pipelines

• Built entire notification system ground-up for consumer apps

• Developed back-end for text-to-video generation tool

Recommendation AlgorithmsReal-Time MLNotification SystemText-to-Video

Machine Learning Engineer

TikTok

Feb 2021 - Aug 2022 · 1 yr 7 mos

Singapore, Singapore · On-site

• Upgraded user-level age prediction model — boosted DAU, social engagements & stay duration at billion-user scale

• Improved recommendation algorithms for content diversity & quality

• Simplified video filtering/boosting strategies, significantly reducing latency

• Enhanced push notification system with optimized sending time strategies

Billion-User ScaleRecommendation AlgoDAU GrowthLatency OptimizationPush Notifications

Data & Applied Scientist

Microsoft

Jul 2019 - Dec 2020 · 1 yr 6 mos

Bengaluru, Karnataka, India · On-site

• Built algorithms to extract relevant info from web pages for Search Ads Annotations

• Shipped end-to-end at Bing.com scale across multiple markets

• Pioneered incorporating visual cues into search ads — granted a USPTO patent

Bing-Scale SystemsSearch AdsUSPTO PatentInformation Extraction

Master of Technology

IIT Bombay

Duration:2017 - 2019

Department:Computer Science & Engineering

Location:Mumbai, Maharashtra, India

Advisor:Prof. Ganesh Ramakrishnan

Specialization

Machine LearningNatural Language ProcessingQuestion Answering SystemResearch

Published Research

"Vocabulary Matters: A Simple yet Effective Approach to Paragraph-level Question Generation"

Authors: Kumar, V., Joshi, M., Ramakrishnan, G., & Li, Y.F.

Conference: AACL-IJCNLP 2020

Achieved 6-point BLEU-4 improvement (16.38 → 22.62) in paragraph-level question generation

Stay in the loop

Get notified when I publish new projects and insights

No spam. Unsubscribe anytime.