Hi, my name is

Bidisha Das

Senior Software Engineer · Backend & Platform Engineering @ Salesforce

I build secure, scalable backend systems and the platforms that let AI agents act safely inside enterprise software.

I’m a Senior Member of Technical Staff at Salesforce, working across enterprise access management, Java and Python backend services, AI agent orchestration, MCP integrations, and permission-aware tool execution with human approval. See the  Experience section for details.

My engineering interests are distributed systems, PostgreSQL, developer infrastructure, and reliable execution layers for AI agents. Before Salesforce I worked on playback reliability at  Disney+ Hotstar, fintech journeys at  MakeMyTrip, and platform tooling at  AutoRABIT. I also contribute to  Shaka Player, Google’s open-source DASH/HLS player.

About Me

I’m a backend and platform engineer with six years of experience shipping production software, most recently on enterprise authorization and AI agent infrastructure at Salesforce.

Most of my work is the execution layer around the model: turning a natural-language request into a plan grounded in real enterprise metadata, then running it through tools that respect the requester’s permissions and stop for human approval when they should. Authorization decisions live in deterministic code, not in a prompt.

I started on the frontend (design systems, performance work, and video playback), and that still shapes how I build platforms other engineers can use. These days I spend most of my time on backend services, PostgreSQL, and making agent workflows testable and safe to operate.

Technologies I work with regularly:

  • Java
  • Spring Boot
  • Python
  • TypeScript
  • PostgreSQL / SQL
  • REST APIs
  • MCP
  • LangGraph
  • RBAC & access control
  • Docker
  • Kubernetes
  • AWS
  • GitHub Actions
  • React
Bidisha Das

Where I've Worked

Senior Member of Technical Staff @ Salesforce

Bengaluru, India

Dec 2024 — Present

Backend and AI-platform work for enterprise access management: Java and Python services, metadata-grounded agent workflows, MCP integrations, and permission-aware execution behind human approval.

  • Java
  • Spring Boot
  • Python
  • TypeScript
  • PostgreSQL
  • MCP
  • REST APIs
  • Metadata & Tooling APIs
  • RBAC
  • AWS
  • Docker
  • Kubernetes
  • Splunk
  • CI/CD
  • Designed, built and deployed five production AI agent workflows for enterprise access management, cutting a process that took ~60 days of manual work to ~5 days of self-serve execution (~92% faster).
  • Extended the in-house Agentforce platform with a retrieval-to-action layer: natural-language requests retrieve versioned Skills, enterprise metadata grounds the plan, and permission-aware tools execute structured operations. 10+ teams now self-serve on it.
  • Authored the MCP server that exposes eight enterprise capabilities to agents, with structured tool contracts, explicit failure semantics and permission-aware execution.
  • Shipped a Slack agent adopted by 150+ people across 8 teams, mostly non-engineers. Privileged operations run only under the requester’s own permissions and behind human approval.
  • Built governed Metadata API and Tooling API integrations across roles, groups, queues, sharing rules and access policies, with schema validation and authorization decisions kept in deterministic code, reducing misconfigured access requests by ~35%.
  • Built Java services on PostgreSQL for role and queue access summaries, restriction rules and public-group membership queries, plus field-level security checks.
  • Built the agent evaluation and regression framework: behavioral checks on retrieval, tool selection and permissions, an LLM judge with a written rubric and pass thresholds, and permission safety cases that must pass on every change.
  • Migrated five workflows from a legacy V1 retrieval-driven architecture to the redesigned V2 platform, consolidating 20+ prompts, permissions, tool definitions and structured outputs.
  • Deployed and operated agent services on AWS with Docker and Kubernetes; event-driven integrations decoupled workflows from downstream systems, improving throughput by ~40%.
  • Mentor 5+ engineers and contractors through code and design reviews.

Software Development Engineer - II @ Disney+ Hotstar

Bengaluru, Karnataka, India · Remote

Jan 2023 — Dec 2024

Player team, web and living-room (TV) platforms: playback reliability, startup performance and session analytics for a large-scale streaming service.

  • JavaScript
  • TypeScript
  • Shaka Player
  • DASH / HLS
  • React
  • Next.js
  • Web Workers
  • Microfrontends
  • Reduced player failure rate by ~70% by matching recovery to each error class: bounded auto-retry, failover to a secondary source and variant blacklisting, without restarting playback.
  • Fixed session analytics that counted retries as new sessions by redesigning playback-session identity, and defined startup, rebuffer and failure metrics with Product and Data Science.
  • Cut load time by ~10% for 50M+ users with segment prefetching and a concurrency-capped preload manager that parses the manifest while DRM initializes in parallel.
  • Designed a web-worker-based playback reporting API for real-time error tracking.
  • Implemented caption sync for stitched ads on Apple HLS, enabling closed captions across 50+ countries.
  • Upstream Shaka Player contributions: subtitle-role fixes for HLS (#5357) and DASH (#10441), both merged.

Senior Software Engineer @ MakeMyTrip

Bengaluru, Karnataka, India · Hybrid

Jun 2022 — Jan 2023

Contributed to TripMoney, building web and mobile fintech solutions across credit, loan, and forex journeys.

  • React
  • Redux Thunk
  • Redux Saga
  • Web Components
  • Loadable Components
  • Webpack
  • Azure
  • Nginx
  • Charles Proxy
  • Delivered travel and personal loan journeys using React, Redux Thunk, and Web Components.
  • Built and integrated APIs with MVVM architecture, hosted on Azure and blob storage.
  • Led frontend performance improvements — reduced LCP, FCP, and TTFB across devices.
  • Cut bundle size using code-splitting, image compression, sprite sheets, and Webpack optimizations.
  • Revamped the Forex Card experience across desktop and mWeb using Redux Saga and Loadable Components.

Software Developer @ AutoRABIT

Hyderabad, Telangana, India · On-site

Sep 2020 — Jun 2022

Platform tooling, integrations and reusable UI systems for DevOps and CRM release-automation products.

  • Java
  • Spring Boot
  • OData v4
  • AWS
  • Docker
  • Angular
  • NgRx
  • Vue
  • GraphQL
  • Web Components
  • Built a connector that loads external CSV data into Salesforce using Java, Spring Boot and Apache Olingo (OData v4), hosted on AWS EC2 and S3.
  • Built the organisation-wide design system (Angular, PrimeNG, NgRx, Web Components) in a micro-frontend setup, used as the single source of truth for UI implementation.
  • Delivered a Configure-Price-Quote (CPQ) application that contributed to ~$2M in revenue.
  • Applied PWA techniques (cache busting, lazy loading, webpack tuning) to the UI layer deployed on AWS.

Selected Projects

  • Featured Project

    ResolveFlow

    A LangGraph agent that diagnoses GitHub issues with cited evidence and never writes to GitHub without an independent review and explicit human approval. The approval pause is a LangGraph interrupt() checkpointed in PostgreSQL and resumable from a later request. Execution refuses to run without a recorded approval, the only write tools are comment and label, and writes run under the signed-in user’s own GitHub token with a retry-safe write ledger. A 13-check safety-gate suite runs in GitHub Actions on every pull request. Try the  live demo.

    • LangGraph
    • FastAPI
    • PostgreSQL
    • Pinecone
    • OpenAI
    • React
    • GitHub Actions
    ResolveFlow — pending human approval screen

  • Featured Project

    LangChain Documentation Helper

    A retrieval-augmented generation (RAG) chatbot that answers questions about the LangChain docs, built with a Streamlit chat UI on top of a LangGraph tool-calling agent that retrieves grounded context from a Pinecone vector store. Part of my  LangChain deep-dive repo. Try the  live demo.

    • LangChain
    • LangGraph
    • Pinecone
    • OpenAI
    • Streamlit
    LangChain Documentation Helper preview

Earlier Frontend Projects

  • Frontend Project

    Google Clone

    A clone of Google with search facilities and a random image search, built with Next.js.

    • Next.js
    • React
    • Tailwind CSS
    Google Clone

  • Frontend Project

    FlyM

    This project provides the feature of flight booking to the user. The user can book flights, and filter based on location, duration, price of flight, airline and whether the flight has one stop or is it non stop. This project is available on  FlyM.

    • React
    • Redux
    • Thunk
    • Node.js
    • MongoDB
    • Express.js
    • Material UI
    • Axios
    FlyM flight booking app

  • Frontend Project

    Slack Clone

    This project aims to provide a clone of slack. Individual messages, chat rooms and creation of channel are some of the features provided in this. The project is available on  Slack Clone.

    • VS Code
    • React
    • Firebase
    • Material UI
    • Functional Components
    Slack Clone

  • Frontend Project

    Jira Clone

    This project aims to provide a clone of jira dashboard. It is enriched with features like user selection, filter based on keyword, member addition, issue edition and so on. The project is available onJira Clone

    • VS Code
    • Angular
    • Spring Boot
    • MongoDB Atlas
    • RxJs
    • Bootstrap
    • PrimeNG
    Jira Clone

  • Frontend Project

    CovidTracker

    The project aims to show a statistical record in tabular as well as graphical format of COVID-19 effected patients per country. It shows the record of affected, deceased and recovered patients, available onCovidTracker

    • VS Code
    • Angular
    • RxJs
    • Bootstrap
    • PrimeNG
    Covid-Tracker

  • Frontend Project

    Meals Marathon

    A web app that mocks food-ordering app, and deals with the ordering of food items, adding them on cart, and handling cart data using state management, available at  ReactMeals  

    • React
    • Firebase
    • VS Code
    • Redux
    Meals Marathon food ordering app

  • Frontend Project

    Git Guide

    An application that shows the summary of a github user, displaying their public repositories, weekly contribution - commits, addition, deletion, avaialble at  GitGuide  It also handles error scenarios and loading state.

    • Angular
    • Github
    • VS Code
    • NgRx
    Git-Guide

Other Noteworthy Projects

Lastly!

Get In Touch

I’m happy to talk about backend and platform engineering, authorization, or building reliable execution layers for AI agents. Email is the fastest way to reach me; I’m also on LinkedIn.

Say Hello