Headline

AI Engineering Leader & Architect | ex-Draup Data Science Manager | Building Sovereign AI & High-Throughput Inference Systems | IIT Delhi

About

I architect AI systems that solve high-gravity problems at an internet-scale. My engineering philosophy is built on logical rigor and high agency—prioritizing system physics and performance over high-level abstractions. Whether it’s managing 10+ concurrent production systems or optimizing the inference layer, my focus is on building "Sovereign AI" that is both explainable and efficient. The Track Record: Scalability & Leadership At Draup, I transitioned from an individual contributor to a Data Science Manager, leading a team of 6 engineers with a 100% retention rate over 2+ years. - Extreme Scale: Architected systems processing 250M+ profiles and 100M+ Job Descriptions. - Inference Engineering: Optimized high-throughput engines using vLLM with custom C++ GLIBCXX patches, achieving 50+ skill complexity scores per second and a 40% reduction in costs. - Algorithmic Rigor: Reduced prediction error from 25% to 5% in global compensation models through synthetic imputation and latent curve reconstruction. The Current Phase: Strategic Sabbatical I am currently on a Strategic Sabbatical in which my time includes structured daily study blocks to evolve my technical stack. This is an intentional "Deep Work" period focused on: -SSMs & State-of-the-Art Architectures: Researching Mamba, Zamba, and Graph-RAG for high-fidelity reasoning. - Agentic Orchestration: Refining Multi-Agent frameworks using LangGraph and Leiden Clustering for complex clinical and talent intelligence. - System Efficiency: Pushing the boundaries of 4-bit quantization and high-TPS local inference. I value direct-approach communication and am looking to connect with teams building at the intersection of mathematical rigor and scalable AI infrastructure. Areas of Deep Interest: GNNs, Multi-Agent Systems, High-Throughput Inference, and Technical Product Roadmapping.

Location

Bengaluru

AI stack

anthropic apis

cursor

gemini

langgraph

langsmith

Unstructured

Work experience

Professional development at Career Break, Apr 2026 - Present · 5 mos, Bengaluru, Karnataka, India, Currently undertaking a disciplined, self-designed sabbatical focused on technical evolution and sovereign AI research. I have implemented a rigorous daily protocol designed to prioritize long-term logical growth over short-term benefits. Current Technical Research & Development: - Daily study blocks focused on mastering advanced machine learning architectures, including State Space Models (SSMs/Mamba) and GNN-augmented Knowledge Graphs. - Developing high-throughput inference protocols utilizing vLLM with custom C++ library patches to optimize token generation at scale. - Refining Graph-RAG implementations for complex reasoning tasks, specifically focusing on "Stubborn Search" recursive protocols.

Data Science Manager at Draup, Oct 2025 - Mar 2026 · 6 mos, On-site, - Led a high-performance team of 6 Data Scientists, overseeing the end-to-end lifecycle of 10+ concurrent DS/ ML projects from conceptualization to production deployment. Mentored 3 junior Data Scientists to promotion. - Owned hiring for the team; built a 6-person roster with 100% retention over 2+ years.

Data Analyst 4 at Draup, Feb 2025 - Sep 2025 · 8 mos, On-site, - Skill Intelligence Architect (SIA): Designed an LLM-based system to quantify skill complexity for 100M+ Job Descriptions. Optimized inference to process 50+ skill complexity scores per second using vLLM, achieving 94% accuracy on benchmarked skill classifications (reduced processing cost by 40%). Developed a privacycompliant "clean-room" distillation pipeline that leveraged an intermediate auditor model to generate verified synthetic training data, effectively decoupling public model weights from proprietary enterprise data. - Candidate Recommendation Model: Architected a high-precision candidate recommendation engine utilizing LLM entity extraction and semantic embedding similarity to bridge the gap between 100M+ JDs and 250M+ profiles. Engineered a multi-stage scoring framework incorporating skill intersection, talent flow analytics, and a custom power decay function for experience proximity weighting. Optimized retrieval through multi-level Elasticsearch filtering & weighted recommendation logic to deliver high-signal hiring leads; 30% increase in user engagement.

Data Analyst 3 at Draup, Feb 2023 - Feb 2025 · 2 yrs 1 mo, Hybrid, - Basepay Estimation Model: Built a unified compensation engine by fusing fragmented salary data with profile history to reconstruct cost-vs-experience curves, and engineered a synthetic imputation model to predict benchmarks for sparse data intersections; reduced prediction error from 25% to 5%. Developed a unified "Persona Valuation Model," assigning granular weights to multi-dimensional filters (skills, education, experience, etc.) to create a consistent, attribute-driven predictive framework for global salary estimation. - Org Structure Construction: Constructed organizational hierarchies for 50K+ companies using FAISS-based embedding search and a multi-factor scoring heuristic—integrating seniority deltas, business functions, and location proximity; 80% accuracy on manager-report matching; automated reporting for 10+ clients.

Career transition at Career Break, Oct 2022 - Jan 2023 · 4 mos, Bengaluru, Karnataka, India, I am currently preparing myself to become a Data scientist/Financial Analyst during this current career transition phase.

Senior Business Analyst at RedSeer, Mar 2022 - Sep 2022 · 7 mos

Program Manager-Special Projects at Dr. Reddy's Laboratories, Nov 2021 - Feb 2022 · 4 mos, Baddi, Himachal Pradesh, India

Technical Trainee, Young Leadership Development Programme at Dr. Reddy's Laboratories, Aug 2020 - Oct 2021 · 1 yr 3 mos, Hyderabad, Telangana, India

Education

Bachelor of Technology, Chemical Engineering, Indian Institute of Technology, Delhi

Public links

github: https://github.com/PiyushVerma18