Data Science Consultant @ ZS | Building multi-agent GenAI systems and the evaluation infra that keeps them honest in production

Bangalore Urban, Karnataka, India

On-site

On-site






Bachelor's degree, Electrical, Electronics and Communications Engineering
National institute of technology kurukshetra
Join Sakshi and 10k+ other professionals on
Tal is the app where bosses meet engineers & builders directly, without middlemen
Try Tal BossData Science Consultant @ ZS | Building multi-agent GenAI systems and the evaluation infra that keeps them honest in production
I build GenAI systems that run in production, the ones sales reps at Fortune 500 pharma open every morning. Over 6 years at ZS my work has moved from classical ML (a feature-engineering system adopted firm-wide) to multi-agent GenAI platforms, owning them from architecture through deployment. A few things I’ve built: – A multi-agent platform on a modular agent runtime (planner–executor orchestration, tool integration, retries, observability) that cut per-use-case integration from 3 weeks to under 1 – A Next Best Action engine that lifted content engagement 20% across 200+ sales reps – A multilingual pre-call pipeline across 6 languages with 80% adoption within two weeks Here’s what I’ve learned building these: GenAI systems rarely fail loudly. They fail quietly, retrieval degrades, a prompt tweak silently breaks three downstream cases, the model gets confidently wrong in a language nobody on the team speaks. The demo still works. Nobody notices for a month. So I spend as much time on the evaluation harness as on the system itself. It’s also what I build outside work, RAGBench (a RAG evaluation framework), MemoryOS (tiered agent memory), and a multi-stage LLM ranking pipeline (Hybrid RRF → LambdaMART → LLM re-ranker → MMR). Looking to bring this hands-on experience to Applied AI, AI/ML Engineer, and Applied Scientist roles — wherever the job is turning a capable model into a system real users trust.
Bangalore Urban
anthropic apis
claude code
gemini
github copilot
lovable
n8n
Advanced Data Science Consultant at ZS, Jan 2026 - Present · 11 mos, Bangalore Urban, Karnataka, India, Building multi-agent GenAI platforms and LLM systems for Fortune 500 pharma commercial teams — systems that run in production. Agentic AI Platform for Sales Enablement Content: • Designed and built a modular multi-agent platform using DAG orchestration, planner–executor separation, and hybrid retrieval to automate multi-step enterprise sales workflows; drove implementation with client engineering teams across 5+ business use cases. • Partnered with client engineering teams to design and deploy across 5+ business use cases, translating business requirements into solution architecture • Established evaluation pipelines for agent accuracy, robustness, and output quality, enabling faster iteration during development and a structured improvement process for production use. • Shipped a Next Best Action engine that guides sales reps on which content to send to which customer, through which channel, and at what cadence; lifted content engagement 20% across 200+ sales reps. GenAI Platform for Sales Pre-Call Planning & Prioritization: • Developed an LLM pipeline converting structured KPIs into prioritized pre-call summaries across 6 languages; cut runtime from 4+ hours to 15 minutes via PySpark parallelization on AWS SageMaker. • Engineered prompt pipelines with prioritization logic and language-specific evaluation; built a multi-layer pre-deployment evaluator checking hallucination, causal overreach, bias, and structural consistency. • Ran UAT and adoption workshops with sales reps across 3 markets, iterating with business stakeholders to ensure outputs met their bar; driving 80% adoption within 2 weeks.
Advanced Data Science Associate Consultant at ZS, Jan 2023 - Dec 2025 · 3 yrs, On-site, Wayfinder — Representation Learning for Customer Segmentation: • Built Wayfinder, a customer segmentation engine converting raw transactional histories into dense per-customer behavioral embeddings encoding event type, temporal patterns, and frequency. • Trained auto encoders on transaction sequences to learn compact latent representations, concatenated them with static demographic features, and clustered the combined vectors to derive segments. • Surfaced distinct customer archetypes invisible to rule-based segmentation, reshaping clients' market segmentation and targeting strategy.
Advanced Data Science Associate at ZS, Aug 2020 - Jan 2023 · 2 yrs 6 mos, On-site, • Built an automated feature engineering tool that used generative algorithms to score and select aggregated features (prevalence, ARIMA, Gini) from transactional event-level data. • Generated id-level feature sets (recency, frequency, and related aggregations) and deployed the solution on AWS SageMaker. • Adopted across multiple teams firm-wide, significantly cutting down manual feature engineering effort.
Bachelor's degree, Electrical, Electronics and Communications Engineering, National institute of technology kurukshetra
github: https://github.com/sakshi-2606
mindgraph-bridge-38vy.vercel.app: https://mindgraph-bridge-38vy.vercel.app/