Backend & AI engineer. Built RAGs, AI agents, SaaS platforms & voice bots. TypeScript, Python, LLMs. I ship fast.

github

Shryesth Pandey  • Peerlist

Created a real-time dashboard aggregating feedback from 60+ news sites, epapers, and RSS feeds, providing a comprehensive view of public opinion for Govt. of India - Automated data acquisition using Tesseract (OCR), Beautiful Soup, and RSS Feeds. - Employed Zero-Shot Classification and BART Large Language Model for accurate sentiment analysis. Implemented departmental categorization for targeted response and improved accountability. Impact: - Enhanced situational awareness, enabling faster and informed decision-making. - Improved response efficiency, reducing crisis escalation through quicker identification and handling of negative news. - Increased departmental engagement by streamlining communication and fostering greater accountability.

360 News Feedback

A voice AI agent that autonomously makes phone calls to gather customer feedback — conducting natural conversations, transcribing responses, and storing structured data for analysis. ● Built on Pipecat (voice AI pipeline framework) integrated with Twilio for outbound phone call management — handling call initiation, audio streaming, and session lifecycle. ● Engineered real-time STT/TTS pipeline enabling natural back-and-forth conversation flow with customers, with dynamic prompt injection based on feedback context. ● Stored conversation flags, full transcripts, and audio recordings in MongoDB with audio files archived to AWS S3 for durable, queryable storage. ● Resolved breaking API changes in Pipecat after a version upgrade under time pressure — debugging against library source code and Claude-assisted function reconstruction. ● Designed for business use cases: post-purchase feedback, NPS surveys, service quality assessments. Tech stack: Python, Pipecat, Twilio, MongoDB, AWS S3, OpenAI (STT/TTS), Node.js

AI Calling Agent — Voice AI for Business Feed

AnkFin is a personal finance SaaS platform providing actionable financial insights across bank accounts, credit cards, and loans — helping users save smarter, spend wisely, and hit their financial goals. ● Built Node.js + Python backend with Next.js frontend, integrating Plaid (OAuth, Transactions, Webhooks) for real-time bank data aggregation across multiple account types. ● Designed PostgreSQL schema with normalized transaction tables, materialized views for analytics dashboards, and B-tree indexes reducing query time from 2.1s to 120ms. ● Built OCR + LLM pipeline (Tesseract + GPT-4 Vision) for receipt ingestion — acting as an ETL layer from raw images to structured transactions, cutting manual expense tracking time by 95%. ● Integrated Stripe (Payments, Subscriptions) and Unit Banking (ACH, Savings Accounts) for full financial workflow coverage. ● Implemented JWT rotation, bcrypt password hashing, audit logging, and S3 server-side encryption. ● Docker-based CI/CD with automated Prisma schema migrations and Redis-based job queuing. Tech stack: Node.js, Python, Next.js, PostgreSQL, Redis, Docker, Prisma ORM, Plaid API, Stripe, Unit Banking, AWS S3, JWT, GitHub Actions

AnkFin — Personal Finance SaaS

Designed and built a production-ready deployment pipeline for RAG (Retrieval-Augmented Generation) chatbots, enabling rapid deployment across multiple client environments. Key highlights: • Built a reusable pipeline to ingest, chunk, and embed documents into a vector store for retrieval. • Supports multiple LLM backends (OpenAI, Anthropic) with configurable prompt templates. • Automated deployment workflow using CI/CD pipelines for zero-downtime releases. • Modular architecture allows swapping embedding models, vector DBs, and LLMs independently. • Used internally at FactSet to power document Q&A and internal knowledge assistants. Tech stack: Python, Node.js, LangChain, OpenAI, Pinecone/pgvector, AWS EC2, Docker

Chatbot Deployment Pipeline — Scalable RAG-Ba

A cloud-hosted fintech backend consolidating banking, expense tracking, and bill management into a single platform — handling 10K+ monthly transactions. ● Designed PostgreSQL schema with normalized transaction tables, materialized views for analytics dashboards, and B-tree indexes reducing query time from 2.1s to 120ms. ● Built OCR + LLM pipeline (Tesseract + GPT-4 Vision) for receipt ingestion, acting as an ETL layer from raw images to structured transactions — cutting manual expense tracking time by 95%. ● Integrated Plaid (OAuth, Transactions, Webhooks), Stripe (Payments, Subscriptions), and Unit Banking (ACH, Savings Accounts). ● Implemented JWT rotation with refresh tokens, bcrypt password hashing, audit logging, and S3 server-side encryption. ● Docker-based CI/CD with automated Prisma schema migrations and Redis-based job queuing. Tech stack: Node.js, PostgreSQL, Redis, Docker, Prisma ORM, Plaid API, Stripe, Unit Banking, AWS S3, JWT, GitHub Actions

Fintech Transaction Platform

Built an AI-powered out-of-office email agent that goes beyond simple auto-replies — it reads incoming emails, understands context, and sends intelligent, personalized responses. Key highlights: • Uses LLM reasoning to classify email urgency and intent before responding. • Generates context-aware replies that reference the content of the original email. • Configurable response rules: escalation paths, delegation notices, and return timelines. • Integrated with Gmail/Outlook via API for seamless inbox management. • Reduces manual follow-ups and ensures no important email goes unacknowledged while away. Tech stack: Python, OpenAI GPT, Gmail API, Node.js, MongoDB

Out-of-Office AI Agent — Intelligent Email Au

Cloud-native AI platform delivering on-demand personalized meditation content based on user mood and preferences, deployed on AWS. ● Developed FastAPI ML services for personalized meditation generation and autonomous research using OpenAI GPT-4 and AWS Polly for text-to-speech audio synthesis. ● Implemented JWT-based authentication, PostgreSQL user management with session tracking, and S3 audio caching — reducing repeat request latency by 78%. ● Deployed on AWS EC2 with Docker containerization, Nginx reverse proxy, PM2 process management, and GitHub Actions CI/CD pipeline with automated database migrations. ● Designed async worker architecture with Celery for non-blocking meditation content generation. Tech stack: Python, FastAPI, OpenAI API, AWS (EC2, S3, Polly), PostgreSQL, Docker, Nginx, Celery, JWT, GitHub Actions

Personalized Meditation Platform

ProfileScout.io is a lead generation and outreach SaaS tool that helps recruiters and sales teams find, qualify, and contact candidates or prospects at scale using AI-powered search and automation. ● Built Google X-Ray search engine to surface targeted LinkedIn profiles and candidate data without relying on LinkedIn’s API — enabling precise Boolean-style candidate discovery. ● Integrated Hunter.io API for automated email discovery and verification across sourced profiles. ● Built Gmail integration for bulk personalized email outreach with attachment support — automating the full send workflow from prospecting to contact. ● Designed scalable backend with Node.js and PostgreSQL to manage prospect pipelines, track email statuses, and store outreach history per campaign. ● Built dashboard for managing campaigns, tracking open/reply rates, and filtering prospects by role, company, and location. Tech stack: Node.js, PostgreSQL, Next.js, Hunter.io API, Gmail API, Google Search (X-Ray), AWS

ProfileScout.io — AI-Powered Recruiter Outrea

Developed "Project Mythos" a 2D RPG game based on Ramayana in Unity Game Engine, bringing India's epic scripture to life for a wider audience. - Led a team in character design, Unity scripting, and level design to create an immersive experience. - Demonstrated technical expertise in C# scripting, character design, and level design within the Unity 3D engine. Recognition: - Awarded ₹25,000 cash prize from the Government of India for cultural innovation

Project Mythos

Developed "Project Mythos" a 2D RPG game based on Ramayana in Unity Game Engine, bringing India's epic scripture to life for a wider audience. - Led a team in character design, Unity scripting, and level design to create an immersive experience. Recognition: - Awarded ₹25,000 cash prize from the Government of India for cultural innovation - Demonstrated technical expertise in C# scripting, character design, and level design within the Unity 3D engine.

Project Mythos: Winner of Toycathon 2021 Digi

An end-to-end Retrieval-Augmented Generation (RAG) system for enterprise document intelligence. Ingests multi-format documents (PDF, DOCX, Excel), chunks and embeds them using OpenAI embeddings, stores in a FAISS/pgvector vector database, and answers natural language queries with source citations. Key features: • Multi-format document ingestion with automated chunking and metadata extraction • Semantic search with hybrid keyword + vector retrieval • Multi-LLM routing (OpenAI GPT-4, Gemini, DeepSeek R1) for optimal cost/quality tradeoffs • CI/CD pipeline via GitHub Actions for automated document updates to the knowledge base • REST API backend built with FastAPI • Achieved 85% retrieval accuracy in production deployment at FactSet Tech stack: Python, FastAPI, LangChain, FAISS, pgvector, PostgreSQL, OpenAI API, GitHub Actions, Docker

RAG Document Intelligence Pipeline

Spearheaded the development of a groundbreaking text-to-video generator, transforming government press releases into informative videos. - Championed modular coding practices for code maintainability, reusability, and efficient collaboration. - Architected a robust data pipeline, utilizing web scraping, BART models for text summarization, and Meta's language models for multilingual translation. - Integrated Meta's Seamless Communication Library for high-quality text-to-speech conversion. - Developed a user-friendly web interface with Gradio for seamless interaction. - Utilized WSL Ubuntu and Git for streamlined development workflows.