Subhranshu P. profile photo 1 on Tal

Subhranshu P.

Software Engineer | GenAI, Agentic RAG, Multi-Agent Systems | LangGraph, Kubernetes, Python | Scaled AI search across 2.9M+ docs | MS CS @ Georgia Tech

Delhi

2+ years of work experience

Everstar
Jul 20241 y 11 moJun 2026

NY, USA (remote)

Founding software engineer at a $4M-seed startup bringing GenAI to nuclear compliance. Built the platform end to end; the work helped land partnerships with Microsoft, NVIDIA, and Accenture. - Cut compliance drafting from weeks to under an hour by architecting a multi-agent LLM workflow in LangGraph (planner, draft composer, drift analysis agents) with citation-accuracy validation and hallucination guards - Accelerated retrieval by 90% across 2.9M+ NRC/DOE/DOD documents by building a hybrid semantic search platform: OpenSearch dense/sparse vector retrieval, Cohere reranking, metadata filtering - Scaled ingestion to millions of PDFs with Airflow-on-Kubernetes pipelines and a CUDA-accelerated FastAPI parsing/OCR microservice (A100 VM cluster, autoscaling, model caching), with contextual chunking preserving cross-page context - Led a zero-downtime production migration from GCP to Azure AKS, codifying the platform (Airflow, OpenSearch, Percona MongoDB) as Terraform modules deployed via ArgoCD GitOps; cut deploys to a single merge with GitHub Actions CI/CD - Ensured reliability of production AI systems with LLM evaluation and end-to-end tracing in LangSmith plus OpenTelemetry, Prometheus, and Grafana monitoring Stack: Python, TypeScript, FastAPI, LangChain, LangGraph, LangSmith, OpenSearch, MongoDB, PostgreSQL, Neo4j, Airflow, Kubernetes (AKS), Terraform, ArgoCD, Azure, GCP
ethica
Feb 20244 mosMay 2024

Personalised House Search Optimisation using Generative AI ◦ Concept: Developed a groundbreaking solution to enhance house search experience by leveraging Generative AI. Captured user's lifestyle and ideal house descriptions via audio, transcribed, and extracted key parameters using advanced AI algorithms. Utilised these parameters to query thousands of listings in the database, applying user-specific criteria to curate personalised recommendations. ◦ Frontend: Spearheaded development using a range of cutting-edge frontend technologies including NextJS, TailwindCSS, Next-Auth, Framer Motion, Zustand, and Jotai to deliver an intuitive user interface. ◦ Backend: Engineered robust backend architecture employing a diverse set of technologies such as Supabase, Prisma, OpenAI SDK, Anthropic SDK, Perplexity SDK, ScraperAPI, Mapbox, and Assembly AI to handle data processing, analysis, and integration seamlessly. ◦ Impact: Significantly optimised workflow efficiency by streamlining the search process, resulting in a 1.5x improvement in overall productivity.
ethica
Sep 20235 mosJan 2024

Geo-Location Visualisation for House Listings ◦ Concept: Developed a dynamic interface akin to Zillow, integrating geolocation visualization with detailed listing information. Implemented a map-based interface displaying real house listings with corresponding markers. ◦ Frontend: Used NextJS and TailwindCSS to meticulously recreate the user-friendly interface reminiscent of Zillow’s intuitive design, ensuring seamless navigation and information accessibility. ◦ Backend: Used Google Maps API to integrate geolocation functionalities seamlessly into the platform, enabling users to visualise property locations effortlessly. ◦ Impact: Optimised query time by 2.5x
CracKube Pvt Ltd
Jul 20233 mosSep 2023

Chennai, Tamil Nadu, India

Built core UI for an edutainment Q&A platform with React, Redux, Firebase, and TailwindCSS
GitHub
@CantBeSubh
YouTubeyoutube.com/@cantbesubh

Subhranshu’s AI Stack

claude
claude
hugging face
hugging face
langgraph
langgraph
langsmith
langsmith
opencode
opencode

Education

Georgia Institute of Technology

Master's degree (Online/Part-time), Computer Science

Georgia Institute of Technology

20262 yrs2028
Vellore Institute of Technology

Bachelor of Technology - BTech, Computer Science

Vellore Institute of Technology

20204 yrs2024

Join Subhranshu and 10k+ other professionals on

Tal Boss

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

Download Tal Boss

Subhranshu P. on Tal

Headline

Software Engineer | GenAI, Agentic RAG, Multi-Agent Systems | LangGraph, Kubernetes, Python | Scaled AI search across 2.9M+ docs | MS CS @ Georgia Tech

About

I build production GenAI systems: agentic RAG, multi-agent LLM workflows, semantic search, and the cloud-native infrastructure to run them at enterprise scale. I was the founding software engineer at Everstar, a NYC startup that raised a $4M seed round to bring GenAI to nuclear compliance. The platform I built end to end helped land partnerships with Microsoft, NVIDIA, and Accenture. Outside work I'm doing an online MS in Computer Science at Georgia Tech, tinkering with my homelab and jailbraking my android phone. Stack: Python, TypeScript, FastAPI, LangChain, LangGraph, LangSmith, OpenSearch, Milvus, MongoDB, PostgreSQL, Airflow, Docker, Kubernetes, Terraform, ArgoCD, Azure, GCP, Next.js, React. Reach me at work.suhbranshu@gmail.com | Portfolio: subhranshu.com | GitHub: github.com/CantBeSubh

Location

Delhi

AI stack

claude

hugging face

langgraph

langsmith

opencode

Work experience

Founding Software Engineer at Everstar, Jul 2024 - Jun 2026 · 1 y 11 mo, NY, USA (remote), Founding software engineer at a $4M-seed startup bringing GenAI to nuclear compliance. Built the platform end to end; the work helped land partnerships with Microsoft, NVIDIA, and Accenture. - Cut compliance drafting from weeks to under an hour by architecting a multi-agent LLM workflow in LangGraph (planner, draft composer, drift analysis agents) with citation-accuracy validation and hallucination guards - Accelerated retrieval by 90% across 2.9M+ NRC/DOE/DOD documents by building a hybrid semantic search platform: OpenSearch dense/sparse vector retrieval, Cohere reranking, metadata filtering - Scaled ingestion to millions of PDFs with Airflow-on-Kubernetes pipelines and a CUDA-accelerated FastAPI parsing/OCR microservice (A100 VM cluster, autoscaling, model caching), with contextual chunking preserving cross-page context - Led a zero-downtime production migration from GCP to Azure AKS, codifying the platform (Airflow, OpenSearch, Percona MongoDB) as Terraform modules deployed via ArgoCD GitOps; cut deploys to a single merge with GitHub Actions CI/CD - Ensured reliability of production AI systems with LLM evaluation and end-to-end tracing in LangSmith plus OpenTelemetry, Prometheus, and Grafana monitoring Stack: Python, TypeScript, FastAPI, LangChain, LangGraph, LangSmith, OpenSearch, MongoDB, PostgreSQL, Neo4j, Airflow, Kubernetes (AKS), Terraform, ArgoCD, Azure, GCP

Product Engineer at ethica, Feb 2024 - May 2024 · 4 mos, Personalised House Search Optimisation using Generative AI ◦ Concept: Developed a groundbreaking solution to enhance house search experience by leveraging Generative AI. Captured user's lifestyle and ideal house descriptions via audio, transcribed, and extracted key parameters using advanced AI algorithms. Utilised these parameters to query thousands of listings in the database, applying user-specific criteria to curate personalised recommendations. ◦ Frontend: Spearheaded development using a range of cutting-edge frontend technologies including NextJS, TailwindCSS, Next-Auth, Framer Motion, Zustand, and Jotai to deliver an intuitive user interface. ◦ Backend: Engineered robust backend architecture employing a diverse set of technologies such as Supabase, Prisma, OpenAI SDK, Anthropic SDK, Perplexity SDK, ScraperAPI, Mapbox, and Assembly AI to handle data processing, analysis, and integration seamlessly. ◦ Impact: Significantly optimised workflow efficiency by streamlining the search process, resulting in a 1.5x improvement in overall productivity.

Product Engineering Intern at ethica, Sep 2023 - Jan 2024 · 5 mos, Geo-Location Visualisation for House Listings ◦ Concept: Developed a dynamic interface akin to Zillow, integrating geolocation visualization with detailed listing information. Implemented a map-based interface displaying real house listings with corresponding markers. ◦ Frontend: Used NextJS and TailwindCSS to meticulously recreate the user-friendly interface reminiscent of Zillow’s intuitive design, ensuring seamless navigation and information accessibility. ◦ Backend: Used Google Maps API to integrate geolocation functionalities seamlessly into the platform, enabling users to visualise property locations effortlessly. ◦ Impact: Optimised query time by 2.5x

Frontend Developer Intern at CracKube Pvt Ltd, Jul 2023 - Sep 2023 · 3 mos, Chennai, Tamil Nadu, India, Built core UI for an edutainment Q&A platform with React, Redux, Firebase, and TailwindCSS

Education

Master's degree (Online/Part-time), Computer Science, Georgia Institute of Technology

Bachelor of Technology - BTech, Computer Science, Vellore Institute of Technology

Public links

github: https://github.com/CantBeSubh

youtube.com: https://youtube.com/@cantbesubh