Shikhar Seth

Senior AI Engineer at Knit (2025-06 – Present) AI-native quant-qual market research platform (Series A) • Architected a multi-agent pipeline to transform raw survey data into stakeholder-ready presentations, cutting deck generation from 48 hours to 45 mins; built tool-calling agents for ingestion, insight generation (sandboxed code execution+judge LLM feedback loop), visualization, and PPTX synthesis, with human in the loop refinement

• Built the shared AI infrastructure layer (LLM routing, prompt management, observability, workflow orchestration) adopted across all production agents; designed a context-driven questionnaire generation module that adapts to each client's research objectives, brand voice, and past studies, cutting survey design time by 40%; built all user-facing UIs in React/TypeScript

•

Designed eval framework grading AI-generated insights and decks via rubric-aligned multi-judge LLM consensus; iterated against research benchmarks to reach human-comparable quality, catching regressions before release

• Partnered with market researchers, product, and leadership to align system design with domain expertise and business priorities; led a brown bag series for 15+ team members, improving adoption of emerging AI capabilities

Team Lead (Data Science) at Info Edge (Naukri.com) (2020-07 – 2025-05) India's largest job portal and online classifieds

• Owned ML for the Ed-Tech (Shiksha) vertical; Designed and productionized a neural learning-to-rank search system leveraging transformers, semantic embeddings, and click-through signals, boosting NDCG and CTR
• Shipped a RAG based llama LLM, creating a first-of-its-kind personalized chatbot for education counselling sector; Implemented PEFT & Rope Scaling to significantly enhance inference & user feedback(>85%); Served via vLLM on in-house GPUs to achieve high concurrency across multiple production touchpoints, including Whatsapp
• Designed and deployed two distinct recommender systems serving 15M+ users: a sequential BERT4REC model and a content recommender leveraging novel institute2vec embeddings, evaluated via A/B tests using behavioral cohorts for personalization to drive a 25% lift in CTR
• Developed Hinglish Translator, Indian-accented SpellChecker, and extractive Reviews Summarizer using transformer and fine-tuned BERT embeddings, improving user engagement by 15% across tier-2/3 city searches
• Architected robust MLOps infrastructure using Docker, Kubernetes, and MLflow, enabling automated training, monitoring, and deployment of production ML models with 99.9% uptime and 35% reduced inference latency

Research Intern at Niramai (2020-02 – 2020-05) Early Stage Startup working on radiation free Breast Cancer Detection

• Improved detection sensitivity from 90% to 92% using novel CNN variants, matching mammography performance

Engineering Intern at Bioretics (2018-11 – 2019-06) Academic Startup integrating ML solutions into Industrial Grade Machines

• Built active learning framework for image segmentation reaching peak performance with 53% images, halving cost

Visiting Researcher at Ecole Polytechnique (2019-01 – 2019-05)

• Identified CRNN-auditory pathway parallels through filter visualization; achieved highest grade for thesis novelty

Visiting Researcher at CNRS (2018-05 – 2018-08) French National Centre for Scientific Research

• Improved adversarial robustness of CNNs by 20% using semantic knowledge, work led to a conference publication

Post Graduate Programme in Machine Learning – Plaksha Fellowship (2019-08 – 2020-07)

B.Tech in Computer Science Engineering – Manipal University (2015-08 – 2019-07)