Available for work
Hi, I'm Soham đź‘‹
Building cool stuff with code. Hackathon enthusiast and AI tinkerer. Google SWE Intern 2025.
SP

About

Hey! I'm Soham, and I absolutely love building things. Whether it's crafting AI agents that can actually hold conversations, architecting real-time data pipelines at Google, or hacking together MVPs at 3 AM during hackathons, I'm all in. I've won 10+ hacks and my playground includes full-stack development, machine learning, and distributed systems. I have also interned at Google where I reduced analytics latency by 85%, and before that, I built various AI-powered tools at scale. When I'm not coding, you find me experimenting with digital art or tinkering Genrative AI models.

Work Experience

G

Google

May 2025 - Aug 2025
Software Engineering Intern
Architected a real-time, event-driven data pipeline that reduced analytics latency by 85% for Google Maps content moderation, transforming 15+ hour batch processes into sub-2-hour operations. Engineered a Java service for event notification processing and implemented a C++ transformation module for distributed columnar storage ingestion. Designed privacy-compliant storage schemas with user data wipeout mechanisms. Leveraged Protocol Buffers, RPCs, and Pub/Sub Stubby Gateway for seamless inter-service communication across Google's infrastructure.
C

Cere Labs

Oct 2024 - Apr 2025
Artificial Intelligence Intern
Migrated Robotic Process Automation workflows from UiPath to Microsoft Power Automate, leveraging Large Language Models to convert and adapt platform-specific syntax. Streamlined automation processes across multiple platforms, enhancing workflow efficiency and reducing manual intervention through intelligent code migration and LLM-powered syntax translation.
M

Meresu

May 2024 - Nov 2024
Full Stack Developer
Developed an interview automation platform featuring AI-powered voice calls for candidate assessment, reducing interview time by 60%. Implemented a comprehensive analytics dashboard for visualizing interview metrics, skill gaps, and performance insights. Integrated LLMs for real-time transcript analysis and personalized hiring recommendations. Designed a scalable architecture with PostgreSQL and serverless functions, managing complex data relationships and automating interview scheduling with cron jobs.
G

Genoshi.io

Feb 2024 - May 2024
Software Developer
Led development of a Visual AI RAG Agent Builder for an IIT Delhi-based startup, implementing React Flow for intuitive workflow design. Engineered a Cite2PDF-powered chatbot delivering sub-second query responses with verifiable citations, reducing research time by 80%. Designed and built modern landing pages for Quest and Genoshi.io products with cutting-edge UI/UX. Secured $3,000 prize at Soonami Venturethon, a prestigious global competition for founders and innovative startups.
C

CDAC

Jun 2024 - Feb 2025
Research Intern
Developed a Flowchart Similarity Detection System focusing on structural and textual similarity analysis. Trained a custom YOLOv8 model to detect flowchart shapes and arrows for accurate structural representation. Integrated OCR to extract text from flowcharts and mapped it to corresponding shapes for detailed analysis. Utilized graph-based data structures to model flowcharts, applying graph isomorphism techniques to compare node, edge, spatial, and textual similarities.

Technical Skills

Languages

Python
TypeScript
JavaScript
C++
Java
SQL

Frontend

React
Next.js
React Flow
TailwindCSS
Streamlit

Backend

Node.js
FastAPI
MongoDB
PostgreSQL
Pinecone

AI/ML

LangChain
PyTorch
HuggingFace
OpenCV
SpaCy
NER

DevOps

Docker
Kubernetes
Protocol Buffers
RPCs
Pub/Sub

Tools

Git
AWS
ElevenLabs
Groq
Deepgram
Gemma
Cohere

My Projects

VYOM

A Banking Appointment Management System with MFA authentication, including facial recognition, government ID verification and OTP. Built a multilingual, multimodal chatbot using Vercel AI SDK and Gemini LLM to render context-aware UI components. Implemented fraud prevention with Isolation Forest, fraud detection using graph based algorithms and credit risk assessment with Random Forest, prioritizing high-value customers.

Next.js
FastAPI
SQLite
Python
InsightFace
Vercel AI SDK
Gemini LLM

RE-DACT

An anonymization tool using NER models like SpaCy, Flair and Multilingual RoBERTa for Hindi, along with Regex for sensitive data redaction. Enabled synthetic data generation with BERT and Faker. Integrated OCR, OpenCV for image redaction and XTTS for audio cloning. Supported multiple formats with customizable redaction levels.

SpaCy
Flair
BERT
Regex
Faker
OCR
OpenCV
XTTS
Python
Streamlit

CLEO

An AI-powered sales assistant that generates tailored proposals from uploaded PDFs, supports AI voice calling and includes a chatbot with voice and text input. Integrated bulk email sending and built an interactive admin UI with a Kanban board and analytics. Implemented Retrieval Augmented Generation (RAG) with multi-query retrieval for proposal generation.

Next.js
FastAPI
LangChain
MongoDB
Pinecone
Llama3
Groq
11Labs

AXON

An AI-powered generative UI system fine-tuned on ShadCN components to generate unique themes and React code using ShadCN UI and TailwindCSS. Enabled seamless microapp deployment via text and image prompts, enhancing user experience. Built a marketplace for users to discover, customize, and deploy AI-generated microapps.

Next.js
PostgreSQL
ShadCN
Gemini AI Studio
TailwindCSS

Hackathons

  • M

    Maha Hackathon Challenge 1.0

    The Oberoi, Mumbai, India

    Developed MahaSevak as a unified, AI-first platform to streamline governance in Maharashtra by centralizing services, automating workflows and providing multilingual citizen support. Intelligent modules like MahaSahayak, MahaChat, MahaSamadhan and MahaVerify simplify citizen interactions, accelerate document verification and enhance service accessibility. Predictive and fraud detection engines, MahaDrishti and MahaSecure, empower proactive governance, disaster management and prevent financial leakages.
  • T

    Tech Expo - Trinity 2025

    SVKM's Dwarkadas J. Sanghvi College of Engineering, Mumbai, India

    Developed a Lunar Surface Simulation system using Unity and BitMiracle.LibTiff to visualize Chandrayaan-2 TMC DEM data with realistic terrain and lighting based on solar angles. The solution integrates chunk-based rendering, LOD and floating-point error prevention to achieve real-scale, high-fidelity, real-time lunar visualization.
  • U

    Union Bank of India's iDEA Hackathon

    K. J. Somaiya College of Engineering, Mumbai, India

    Developed VYOM, a secure banking appointment system with facial recognition, MFA, AI-powered multilingual and multimodal chatbot, AI voice calling, ML-driven fraud prevention and credit risk assessment along with graph based fraud detection.
  • H

    Hackniche 3.0

    SVKM's Dwarkadas J. Sanghvi College of Engineering, Mumbai, India

    Developed ShopMart, a multi-stage hybrid recommendation system using GraphQL, ML and GCNs to enhance accuracy, tackle cold start issues and provide personalized, context-aware suggestions—featuring RAG-powered and visual search, loyalty programs, behavioral tracking and business intelligence alerts.
  • S

    Smart India Hackathon

    Indian Institute of Technology Kharagpur (IIT-K), West Bengal, India

    Developed RE-DACT, an anonymization tool using NER models, OCR and OpenCV to redact sensitive data across text, images, audio and video, with synthetic data generation and customizable redaction levels.
  • 1

    100xEngineers Generative AI Buildathon

    Online

    Developed CLEO, an AI-powered sales assistant supporting AI-driven proposal generation with RAG, AI voice calling, voice-enabled chatbot and an interactive admin dashboard with analytics.

Get in Touch

Let's Connect

Want to chat? Just shoot me a DM @X or send me an email below

Email Form