Annanya Bansal

• Building and shipping production-grade portfolio systems — multi-vector RAG pipelines, multi-agent
orchestration (LangGraph), and domain fine-tuned LLMs — completed advanced certifications in agentic AI
evaluation, MCP, and Azure OpenAI.

• Designed and maintained SQL-based data pipelines and preprocessing workflows supporting analytics and
ML use cases across multiple client projects.
• Managed database architecture and executed large-scale data transformations for enterprise clients.

• Architected and delivered AccessIQ — an end-to-end LLM-powered workforce upskilling platform — serving
enterprise clients across finance, HR, and education verticals.
• Engineered an LLM pipeline that auto-generates the top 30 role-specific skills for any industry, analyzes skill
gaps against employee proficiency benchmarks, and recommends personalized reskill journeys, reducing
manual L&D analysis effort by an estimated 60%.
• Designed a dynamic assessment generation module that creates role-specific evaluations via LLM when
client benchmarks are unavailable, expanding platform coverage to 100% of onboarding clients.
• Built semantic search and embedding-based retrieval (FAISS + Pinecone) for course and pathway
recommendations, significantly improving knowledge discovery speed for HR teams.
• Developed structured prompt templates and agent-style workflows ensuring consistent, auditable LLM
outputs for business-critical workforce decisions.
• Monitored LLM performance via LangSmith and optimized inference cost and latency through iterative
prompt engineering and model selection, achieving measurable per-query cost reduction in production.

• Built Call Companion and Sales Coach — AI products deployed within Microsofts environment for real-time
NLP analysis of customer representative call transcripts.
• Engineered an NLP pipeline extracting structured per-representative performance insights: strengths,
coaching gaps, and specific recommendations for better customer handling.
• Automated per-representative feedback generation at scale, reducing manual QA review effort by 40%+
across hundreds of daily calls.
• Refined NLP model performance through preprocessing improvements and feature tuning; automated ETL
workflows using Python and SQL to stabilize production data pipelines.

• Architected and delivered enterprise knowledge assistants and domain-specific chatbots using RAG
architectures, semantic retrieval, and LLM APIs for multiple clients across industries.
• Implemented embedding-based retrieval systems that measurably reduced LLM hallucination rates and
improved response grounding in production deployments.
• Built scalable REST APIs (FastAPI) and interactive AI interfaces serving real end users; developed ML
pipelines for classification, recommendation systems, and analytics dashboards.

• Delivered AI-driven customer intelligence for TATAs retail portfolio — one of Indias largest retail
ecosystems — across multiple product lines and customer segments.
• Built RFM segmentation models identifying high-value customer cohorts, directly informing personalized
coupon and offer generation strategies that improved campaign conversion rates.
• Developed churn prediction models enabling proactive retention campaigns and improving customer lifetime
value across retail clients.
• Automated analytics workflows using Python and SQL, reducing reporting time by 50%+ and enabling faster
campaign decision-making.

• Delivered AI and data analytics strategy for Aam Aadmi Party (AAP) ahead of Punjab and Goa state elections
(2021-22).
• Built predictive models and campaign analytics contributing to AAPs historic Punjab victory — improving
vote share from 29% to ~60% and securing a legislative majority.
• Engineered a real-time call analysis system processing constituency-level outreach calls to identify
candidate favourability and voter sentiment across Punjab, enabling data-driven candidate selection.
• Delivered campaign intelligence for Goa operations where AAP had zero prior presence — analytical insights
contributed to securing ~10% of assembly seats in the debut election.

• Built regression, classification, and time-series models for clients across various industries; delivered EDA,
feature engineering, and actionable business insights.