Soham Chaudhari profile photo 1 on Tal

Soham Chaudhari

AI Developer | Generative AI & LLM Systems | RAG, LangGraph, PyTorch, Multimodal AI | Building Production-Grade AI Applications | AI Intern @ FLIP

Mumbai Metropolitan Region

1+ year of work experience

FLIP
Feb 20265 mosPresent

Remote

- Engineered an 11-node LangGraph orchestration pipeline enabling hybrid retrieval across 9,700+ credit card offers, combining semantic search, SQL filtering, vector embeddings, and web fallback for high-accuracy recommendations. - Designed a PostgreSQL-backed semantic memory layer with async session storage and feedback APIs, cutting token consumption by 60–70% through optimized context injection. - Deployed and maintained scalable AI backend infrastructure on AWS EC2 using Docker Compose, connection pooling, API rate limiting, and health monitoring — ensuring high availability and sub-200ms response times. - Optimized retrieval and inference workflows across the full AI stack, enabling production-scale performance with low latency and fault tolerance.
UNLAWC
Sep 20258 mosApr 2026

Mumbai, Maharashtra, India

- Architected RAG pipelines using LangChain and Pinecone with reranking and self-correction workflows, improving retrieval accuracy by 60% and significantly reducing LLM hallucinations in legal document Q&A. - Reduced AI inference costs by ~80% through optimized retrieval strategies, efficient prompt design, and streamlined deployment — without sacrificing output quality. - Improved LLM response reliability by 31% via structured outputs, evaluation-driven iteration, and systematic prompt engineering on legal domain tasks. - Automated end-to-end legal document analysis workflows, cutting manual review effort by ~50% and enabling faster turnaround for legal teams.
UNLAWC
Sep 20256 mosFeb 2026

Deployed end-to-end Legal Operations AI agents using LLMs, reducing manual document review time by 50%. Optimized LLM performance through prompt engineering, improving contextual accuracy by 31%. Architected RAG pipelines with Mistral AI, LangChain, and Pinecone vector database, increasing retrieval accuracy by 60% and eliminating hallucinations.
SR Counselling India
Dec 202410 mosSep 2025

Mumbai, Maharashtra, India

- Built and deployed a production ML recommendation engine processing 10,000+ daily interactions at 95% model accuracy, directly increasing user engagement by 35%. - Developed 15+ REST APIs with Node.js and Express.js, supporting AI workflows and handling 5,000+ concurrent requests with stable performance under load. - Designed an AI-powered VISA training system that reduced operational training costs by ~90%, replacing manual processes with an automated, scalable pipeline. - Built scalable React.js admin dashboards with reusable components, reducing page load times by 60%, and improved Flutter mobile app performance, boosting retention metrics by 35%.

Education

Vivekanand Education Society's Institute Of Technology

Bachelor of Engineering, Computer Engineering

Vivekanand Education Society's Institute Of Technology

Join Soham and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Soham Chaudhari on Tal

Headline

AI Developer | Generative AI & LLM Systems | RAG, LangGraph, PyTorch, Multimodal AI | Building Production-Grade AI Applications | AI Intern @ FLIP

About

I build production-grade AI systems that solve real problems — not demos, not prototypes. As an AI Engineer, I specialize in LLM applications, RAG pipelines, AI agents, and multimodal systems. I care deeply about reliability, cost efficiency, and systems that actually work at scale. At FLIP, I engineered an 11-node LangGraph orchestration pipeline with hybrid retrieval across 9,700+ credit card offers, cutting token consumption by 60–70% through smart memory injection. At UNLAWC, I built Legal-Tech AI systems that reduced inference costs by ~80% and improved retrieval accuracy by 60% using LangChain, Pinecone, and evaluation-driven iteration. Core stack: LLMs · RAG · LangChain & LangGraph · Vector Databases · PyTorch · FastAPI & Node.js · AWS & Docker · Multimodal AI Currently exploring: Agentic AI systems, LLM evaluation frameworks, efficient inference & quantization, and advanced multimodal architectures. Open to: AI Engineer · ML Engineer · Generative AI Engineer · Applied AI · AI Infrastructure roles.

Location

Mumbai Metropolitan Region

Work experience

Artificial Intelligence Intern at FLIP, Feb 2026 - Present · 5 mos, Remote, - Engineered an 11-node LangGraph orchestration pipeline enabling hybrid retrieval across 9,700+ credit card offers, combining semantic search, SQL filtering, vector embeddings, and web fallback for high-accuracy recommendations. - Designed a PostgreSQL-backed semantic memory layer with async session storage and feedback APIs, cutting token consumption by 60–70% through optimized context injection. - Deployed and maintained scalable AI backend infrastructure on AWS EC2 using Docker Compose, connection pooling, API rate limiting, and health monitoring — ensuring high availability and sub-200ms response times. - Optimized retrieval and inference workflows across the full AI stack, enabling production-scale performance with low latency and fault tolerance.

AI Developer at UNLAWC, Sep 2025 - Apr 2026 · 8 mos, Mumbai, Maharashtra, India, - Architected RAG pipelines using LangChain and Pinecone with reranking and self-correction workflows, improving retrieval accuracy by 60% and significantly reducing LLM hallucinations in legal document Q&A. - Reduced AI inference costs by ~80% through optimized retrieval strategies, efficient prompt design, and streamlined deployment — without sacrificing output quality. - Improved LLM response reliability by 31% via structured outputs, evaluation-driven iteration, and systematic prompt engineering on legal domain tasks. - Automated end-to-end legal document analysis workflows, cutting manual review effort by ~50% and enabling faster turnaround for legal teams.

Artificial Intelligence Engineer at UNLAWC, Sep 2025 - Feb 2026 · 6 mos, Deployed end-to-end Legal Operations AI agents using LLMs, reducing manual document review time by 50%. Optimized LLM performance through prompt engineering, improving contextual accuracy by 31%. Architected RAG pipelines with Mistral AI, LangChain, and Pinecone vector database, increasing retrieval accuracy by 60% and eliminating hallucinations.

SDE Intern at SR Counselling India, Dec 2024 - Sep 2025 · 10 mos, Mumbai, Maharashtra, India, - Built and deployed a production ML recommendation engine processing 10,000+ daily interactions at 95% model accuracy, directly increasing user engagement by 35%. - Developed 15+ REST APIs with Node.js and Express.js, supporting AI workflows and handling 5,000+ concurrent requests with stable performance under load. - Designed an AI-powered VISA training system that reduced operational training costs by ~90%, replacing manual processes with an automated, scalable pipeline. - Built scalable React.js admin dashboards with reusable components, reducing page load times by 60%, and improved Flutter mobile app performance, boosting retention metrics by 35%.

Education

Bachelor of Engineering, Computer Engineering, Vivekanand Education Society's Institute Of Technology