Suhotra Dey profile photo 1 on Tal

Suhotra Dey

Immediate Joiner · Software Engineer · Java · Distributed Systems · Event-Driven Architecture · Agentic RAG · LLM Pipelines · MCP · Ex-Founding Engineer @ Buildway.ai

Bengaluru

3+ years of work experience

Buildway
Aug 202511 mosJun 2026

Upper Austria, Austria

- As founding engineer at Buildway , architected and shipped https://alawyer.ai/ end-to-end — agentic legal AI platform live in 4 EU markets (DE, IT, ES, FR), backed by NVIDIA Inception and Google for Startups Accelerator. - Built a hybrid-retrieval Agentic RAG pipeline over a 2.3M+ record legal knowledge base — LlamaIndex embeddings into Weaviate + custom PostgreSQL similarity function, both exposed as Model Context Protocol (MCP) tools for dynamic law retrieval and citation-grounded answers. - Drove the platform's core architectural pivot: collapsed a brittle multi-agent pipeline (context lost across stateless handoffs) into one sandboxed single agent environment in E2B per conversation — eliminating an entire class of context-drift bugs. - Designed multi-model orchestration with automatic provider fallback chains — primary-model degradations route to alternates without user-visible failure. Integrated Langfuse end-to-end for LLM traces, evals, and per-step debugging. - Engineered real-time AI response streaming via Server-Sent Events, cutting time-to-first-token from 3–5s to sub-second, with fault-tolerant session recovery that resumes interrupted generations. - Diagnosed and resolved a production incident: Prisma connection-pool exhaustion under parallel load — root-caused to multiple client instances spinning up per request, applied singleton pattern to enforce a single shared client, and retuned pool size and idle-timeout, restoring stability without downtime. - Integrated Stripe subscription billing with idempotent webhook handling for the full payment lifecycle (charges, refunds, cancellations), and built 100+ office integrations via Google Drive (GCP) and OneDrive (Microsoft Graph API). - Implemented MCP elicitation for mid-session human-in-the-loop feedback — the agent pauses, surfaces a clarifying question, and continues — enabling interactive contract analysis without session restart.
Cognizant
Sep 20222 yrs 11 mosJul 2025

India

- Built and maintained backend services on the center-booking pipeline of a global standardized English-language testing platform — Spring Boot REST endpoints, service-layer logic, and the MySQL query paths underneath. Debugged production incidents on the live pipeline by log-diving and tracing failures to root cause, and tuned query performance on booking data when slow paths surfaced under load across test centers worldwide. - Wrote an Apache Kafka event-driven fan-out job that distributed exam-question payloads to test centers globally when an exam was scheduled — producer logic, consumer handlers, and the message contract between the exam-scheduling service and downstream test-center endpoints. - Owned reliability and observability work on async Apache Kafka consumers: introduced retry handling and dead-letter routing for failure modes that were previously surfacing as opaque upstream timeouts, and tightened logging around the failure paths. - Built internal monitoring tools backed by tuned MySQL query paths and data aggregation, used by the team for diagnostic triage when production incidents surfaced on the booking pipeline. - Worked within full SDLC and Agile delivery on an enterprise Java codebase — Git, Maven, Jenkins-managed CI/CD, and AWS EC2 deployments. Every PR reviewed by senior engineers.

Suhotra’s AI Stack

claude code
claude code
codex
codex
cursor
cursor
mastra
mastra
mcp
mcp
PydanticAI
PydanticAI

Education

Academy of Technology

Bachelor of Technology - BTech, Computer Science Engineering

Academy of Technology

20194 yrs2023

Join Suhotra and 10k+ other professionals on

Tal Boss

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

Download Tal Boss

Suhotra Dey on Tal

Headline

Immediate Joiner · Software Engineer · Java · Distributed Systems · Event-Driven Architecture · Agentic RAG · LLM Pipelines · MCP · Ex-Founding Engineer @ Buildway.ai

About

I build backend systems and agentic AI pipelines that ship to production. As founding engineer at Buildway, I built aLawyer.ai end-to-end — an agentic legal AI platform live across Germany, Italy, Spain, and France, serving paid customers, backed by NVIDIA Inception and Google for Startups Accelerator. That meant owning everything from scratch: → Agentic RAG over 2.3M+ legal records — LlamaIndex embeddings into Weaviate, exposed as Model Context Protocol (MCP) tools with elicitation for citation-grounded retrieval → Multi-model orchestration with automatic provider fallback — primary-model degradations route to alternates without user-visible failure → Reliability pivot: collapsed a brittle multi-agent pipeline (context lost across stateless handoffs) into one sandboxed Claude Code in E2B per conversation — killed an entire class of failures → Prod incident: Prisma connection-pool exhaustion under parallel load — root-caused, retuned pool + idle-timeout, no downtime → SSE streaming cutting time-to-first-token from 3–5s to sub-second, with fault-tolerant session recovery → Stripe billing with idempotent webhooks + 100+ office integrations (Google Drive, OneDrive) Before Buildway, 22 months at Cognizant as a Java backend engineer on a global standardized English-language testing platform: → REST APIs on the test-center booking pipeline — endpoint design, MySQL query optimization, production incident debugging → Apache Kafka event-driven job fanning out exam-question payloads to a global network of test centers — producer/consumer + inter-service contracts → Spring Boot microservices with Spring Cloud (Eureka, Feign, API Gateway), RabbitMQ with DLQs + retries ━━━ What I build ━━━ Backend & Distributed Systems Java · Spring Boot · Spring Cloud · Apache Kafka · RabbitMQ · PostgreSQL · MySQL · Redis · REST APIs · WebSockets · Server-Sent Events · Event-Driven Architecture · Distributed Systems · Microservices · Fault Tolerance · Connection Pool Tuning AI / Agentic Systems Agentic RAG · Retrieval-Augmented Generation (RAG) · RAG Pipelines · Model Context Protocol (MCP) · Large Language Models (LLM) · LLM Pipelines · Prompt Engineering · Multi-Model Orchestration · LlamaIndex · Weaviate · Vector Embeddings · LLM Observability · Langfuse · E2B · Claude Code · OpenAI · Anthropic · Generative AI Languages Java · TypeScript · JavaScript · Node.js · Go · SQL Cloud & DevOps GCP · AWS (EC2) · Docker · Ubuntu Linux · CI/CD · Vercel · Stripe · Microsoft Graph API 📧 deysuhotra@gmail.com · 🔗 github.com/Lucifer4255 · 🌐 suhotra-dey.vercel.app

Location

Bengaluru

AI stack

claude code

codex

cursor

mastra

mcp

PydanticAI

Work experience

Software Engineer at Buildway, Aug 2025 - Jun 2026 · 11 mos, Upper Austria, Austria, - As founding engineer at Buildway , architected and shipped https://alawyer.ai/ end-to-end — agentic legal AI platform live in 4 EU markets (DE, IT, ES, FR), backed by NVIDIA Inception and Google for Startups Accelerator. - Built a hybrid-retrieval Agentic RAG pipeline over a 2.3M+ record legal knowledge base — LlamaIndex embeddings into Weaviate + custom PostgreSQL similarity function, both exposed as Model Context Protocol (MCP) tools for dynamic law retrieval and citation-grounded answers. - Drove the platform's core architectural pivot: collapsed a brittle multi-agent pipeline (context lost across stateless handoffs) into one sandboxed single agent environment in E2B per conversation — eliminating an entire class of context-drift bugs. - Designed multi-model orchestration with automatic provider fallback chains — primary-model degradations route to alternates without user-visible failure. Integrated Langfuse end-to-end for LLM traces, evals, and per-step debugging. - Engineered real-time AI response streaming via Server-Sent Events, cutting time-to-first-token from 3–5s to sub-second, with fault-tolerant session recovery that resumes interrupted generations. - Diagnosed and resolved a production incident: Prisma connection-pool exhaustion under parallel load — root-caused to multiple client instances spinning up per request, applied singleton pattern to enforce a single shared client, and retuned pool size and idle-timeout, restoring stability without downtime. - Integrated Stripe subscription billing with idempotent webhook handling for the full payment lifecycle (charges, refunds, cancellations), and built 100+ office integrations via Google Drive (GCP) and OneDrive (Microsoft Graph API). - Implemented MCP elicitation for mid-session human-in-the-loop feedback — the agent pauses, surfaces a clarifying question, and continues — enabling interactive contract analysis without session restart.

Software Engineer at Cognizant, Sep 2022 - Jul 2025 · 2 yrs 11 mos, India, - Built and maintained backend services on the center-booking pipeline of a global standardized English-language testing platform — Spring Boot REST endpoints, service-layer logic, and the MySQL query paths underneath. Debugged production incidents on the live pipeline by log-diving and tracing failures to root cause, and tuned query performance on booking data when slow paths surfaced under load across test centers worldwide. - Wrote an Apache Kafka event-driven fan-out job that distributed exam-question payloads to test centers globally when an exam was scheduled — producer logic, consumer handlers, and the message contract between the exam-scheduling service and downstream test-center endpoints. - Owned reliability and observability work on async Apache Kafka consumers: introduced retry handling and dead-letter routing for failure modes that were previously surfacing as opaque upstream timeouts, and tightened logging around the failure paths. - Built internal monitoring tools backed by tuned MySQL query paths and data aggregation, used by the team for diagnostic triage when production incidents surfaced on the booking pipeline. - Worked within full SDLC and Agile delivery on an enterprise Java codebase — Git, Maven, Jenkins-managed CI/CD, and AWS EC2 deployments. Every PR reviewed by senior engineers.

Education

Bachelor of Technology - BTech, Computer Science Engineering, Academy of Technology

Public links

github: https://github.com/Lucifer4255

leetcode: https://leetcode.com/deysuhotra