Soham Malakar profile photo 1 on Tal

Soham Malakar

Software Engineer @ ShopOS | Multi-agent Systems Development

Bengaluru

3+ years of work experience

ShopOS
Jul 20251 yr 3 mosPresent

Bengaluru, Karnataka, India

Working on backend & agent infrastructure •Own reliability and performance of our multi-agent systems. Context engineering, harness design, memory. • Led ShopOS Memory. Captures signals from chats, agents and workflows, clusters them offline into knowledge graphs agents read from. Also builds Brand Memory from a brand's URL and docs. • Led the Creative Director Agent. Brand Memory grounded, makes on-brand images, video and scripts. 100+ brands onboarded, shipping months of content in days. • Cut agent token costs 82% by rebuilding context management on sandboxed execution and filesystem-backed context. Also made Performance Marketing analytics much stronger. • Migrated the multi-agent backend from FastAPI + LangGraph to Mastra + Hono, after building the original runtime. • Shipped MarketScribe. RPA pipelines and a rule engine for campaign automation, running with Noon, Namshi and other Middle East retailers. • Built Stripe billing end to end (subscriptions, add-ons, custom plans, coupons) and the platform's multi-tenant notification system.
IMMPROVED
Feb 20241 yr 6 mosJul 2025

Toronto, Ontario, Canada

• Launched an AI-powered mock interview platform adopted by 5 universities, conducting over 1,000 interviews/week with tailored audio/video questions and scenario-based feedback generated from job title, JD, and resume • Built an intelligent resume builder that parsed CVs and job descriptions to auto-generate personalized resumes — used by 2,000+ users, increasing recruiter response rates by 35% • Built a multi-agent AI chatbot using LangGraph.js, fine-tuned on immigration workflows — reduced support tickets by 40% and improved resolution speed by 55% • Architected scalable backends and vector-based RAG pipelines to inject user-uploaded context into conversations, improving AI response accuracy by 60% • Fine-tuned OpenAI models for domain-specific tasks like document parsing and integrated Deepgram for real-time speech assessment • Developed document autofill pipelines to extract data from passport and visa images, cutting immigration form fill-up time by 70%
Cosedge
Dec 20233 mosFeb 2024

Delhi, India

Innovacio Technologies
Jul 20231 yr 9 mosMar 2025

Kolkata, West Bengal, India

• Designed an efficient database structure that connects rehabilitation facilities with categories, images, and geographic details, eliminating redundant data and optimizing query speed by 30%. • Implemented a scraper to compile current rehabilitation data, amassing over 80,000 facilities across the USA and facilitating nationwide search capabilities by category for users. • Fine-tuned OpenAI API to develop engaging templates featuring FAQs and insurance options, leading to over 10,000 daily engagements and driving a 60% increase in traffic. • Collaborated with the marketing team to create a smart search engine capable of generating over 1,000 recommendations and understanding diverse user queries.
GitHub
@codevibesme
LeetCode
@onedayforsure

Soham’s AI Stack

claude
claude
cursor
cursor
langchain
langchain
langgraph
langgraph
langsmith
langsmith
mastra
mastra

Education

Netaji Subhash Engineering College

Bachelor of Technology - BTech, Computer Science and Engineering

Netaji Subhash Engineering College

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 Malakar on Tal

Headline

Software Engineer @ ShopOS | Multi-agent Systems Development

About

I’m a backend engineer who likes working on problems where the line between software and intelligence gets blurry. I’ve spent the last few years building AI products from the ground up, from multi-agent platforms and long-term memory to model routing, knowledge graphs, and the infrastructure that keeps everything running at scale. What I enjoy most is the systems thinking behind it. Designing how agents use context, deciding what they should remember, making different agents work together, and then obsessing over the details that make the whole thing faster, cheaper, and more reliable. I’ve also had the chance to take these ideas into production, across large creative workflows, hundreds of brands, and systems used by other agents and products. I’m curious about where AI is heading, but even more interested in building the infrastructure that gets us there.

Location

Bengaluru

AI stack

claude

cursor

langchain

langgraph

langsmith

mastra

Work experience

Software Engineer at ShopOS, Jul 2025 - Present · 1 yr 3 mos, Bengaluru, Karnataka, India, Working on backend & agent infrastructure •Own reliability and performance of our multi-agent systems. Context engineering, harness design, memory. • Led ShopOS Memory. Captures signals from chats, agents and workflows, clusters them offline into knowledge graphs agents read from. Also builds Brand Memory from a brand's URL and docs. • Led the Creative Director Agent. Brand Memory grounded, makes on-brand images, video and scripts. 100+ brands onboarded, shipping months of content in days. • Cut agent token costs 82% by rebuilding context management on sandboxed execution and filesystem-backed context. Also made Performance Marketing analytics much stronger. • Migrated the multi-agent backend from FastAPI + LangGraph to Mastra + Hono, after building the original runtime. • Shipped MarketScribe. RPA pipelines and a rule engine for campaign automation, running with Noon, Namshi and other Middle East retailers. • Built Stripe billing end to end (subscriptions, add-ons, custom plans, coupons) and the platform's multi-tenant notification system.

Software Developer at IMMPROVED, Feb 2024 - Jul 2025 · 1 yr 6 mos, Toronto, Ontario, Canada, • Launched an AI-powered mock interview platform adopted by 5 universities, conducting over 1,000 interviews/week with tailored audio/video questions and scenario-based feedback generated from job title, JD, and resume • Built an intelligent resume builder that parsed CVs and job descriptions to auto-generate personalized resumes — used by 2,000+ users, increasing recruiter response rates by 35% • Built a multi-agent AI chatbot using LangGraph.js, fine-tuned on immigration workflows — reduced support tickets by 40% and improved resolution speed by 55% • Architected scalable backends and vector-based RAG pipelines to inject user-uploaded context into conversations, improving AI response accuracy by 60% • Fine-tuned OpenAI models for domain-specific tasks like document parsing and integrated Deepgram for real-time speech assessment • Developed document autofill pipelines to extract data from passport and visa images, cutting immigration form fill-up time by 70%

Full-stack Developer at Cosedge, Dec 2023 - Feb 2024 · 3 mos, Delhi, India

Full-stack Developer at Innovacio Technologies, Jul 2023 - Mar 2025 · 1 yr 9 mos, Kolkata, West Bengal, India, • Designed an efficient database structure that connects rehabilitation facilities with categories, images, and geographic details, eliminating redundant data and optimizing query speed by 30%. • Implemented a scraper to compile current rehabilitation data, amassing over 80,000 facilities across the USA and facilitating nationwide search capabilities by category for users. • Fine-tuned OpenAI API to develop engaging templates featuring FAQs and insurance options, leading to over 10,000 daily engagements and driving a 60% increase in traffic. • Collaborated with the marketing team to create a smart search engine capable of generating over 1,000 recommendations and understanding diverse user queries.

Education

Bachelor of Technology - BTech, Computer Science and Engineering, Netaji Subhash Engineering College

Public links

github: https://github.com/codevibesme

leetcode: https://leetcode.com/onedayforsure