Jishnu Saha

About

Experience

Projects

Skills

Certificates

Education

Contact

Blog

About

Experience

Projects

Skills

Certificates

Education

Contact

Blog

Jishnu Saha

AI Engineer · LLM Applications, RAG & Agentic Systems · Python

AI engineer with 6+ years of production software experience in Python, Django and React, now focused on building LLM-powered products. Design and ship RAG pipelines, LangGraph agents and real-time voice AI systems end-to-end: retrieval and orchestration design, async task infrastructure, streaming APIs and AWS deployment. Currently owns the voice-based AI mock interview system at Dualflow AI, a production product that combines streaming speech-to-text, resume-aware LLM question generation, answer scoring and text-to speech. AWS Solutions Architect – Associate certified

View CV

My Experience

Experienced IT software engineer with over 5 years of diverse industry experience, specializing in web app development and AWS DevOps.

Software Engineer (Remote)

Dualflow, San Francisco.

Nov 2025 - Present

Designed an AI workflow that analyzes user resumes to extract skills, experience, and gaps, and automatically generates a personalized mock interview setup (role, topics, and question structure) from the analysis.
Built the voice-based AI mock interview feature end-to-end: real-time speech-to-text over JWT-authenticated WebSockets (Django Channels, Google Cloud Speech-to-Text), plus resume-tailored question generation and answer scoring through LangGraph agents running as async Celery tasks.
Architected the Next.js frontend from scratch, defining the layout system, page routing and transitions, and global state management with Redux Toolkit, and implemented all page designs using TypeScript and Tailwind CSS.

Software Engineer (Remote)

Remotedevs.com, Lithuania.

June 2025 - Sept 2025

Developed pixel-perfect, responsive pages in Next.js from Figma designs, building reusable components with TypeScript and Tailwind CSS.
Integrated pages with backend REST APIs, handling data fetching, loading and error states, and ensuring consistent behavior across devices and browsers.

Software Engineer (Remote)

Graaho Technologies, Dhaka.

Feb 2020 - June 2025

KhaoDao Chatbot: Built a natural-language order-placement chatbot on Facebook Messenger (OpenAI, Python, DynamoDB, AWS Lambda) that let non-technical users complete orders conversationally.
KhaoDao Backend: Built and maintained scalable backend services using Python, Django, AWS Lambda, and PostgreSQL, adding new features while sustaining high system performance.
KhaoDao Frontend: Developed a device-responsive food ordering web platform in ReactJS that improved customer engagement.Live Demo
ZDSL Property Management System: Led end-to-end development of a full property tracking, sales, and account management system using EC2, Django, ReactJS, and PostgreSQL.
AlgoRec Frontend: Delivered the frontend for a recommendation engine platform using ReactJS.Live
Data Pipeline Automation: Automated the KhaoDao restaurant & menu data entry pipeline in Python, removing manual intervention during bulk restaurant onboarding.
Order Automation Bot: Built a Selenium-based bot to automate order placement across multiple third-party online systems.

Software Engineer (Contract, via Graaho)

Renforce, Sydney, Australia.

Oct 2021 – Jan 2023

Built and maintained scheduled Python ETL pipelines (Selenium, BeautifulSoup4) that scraped regulatory websites, normalized the records, and loaded them into a queryable database for downstream analysis.
Implemented OCR-based extraction of structured data from scanned PDF documents and fed the results into the same pipeline.
Developed an internal React web app that automated the team's recurring daily tasks, replacing manual process, spreadsheet tracking.

Software Engineer (Remote - Contract)

Fly No Borders, Dhaka.

Oct 2019 – Jan 2020

Developed a flight fare comparison and booking engine using Python, Selenium, and BeautifulSoup4.

My Projects

Here are some of my projects

Codebase Research Agent

AI-powered agent that automatically researches GitHub repositories and answers implementation-specific questions by exploring source code.

Tech Used:Python, Django, Django REST Framework, LangGraph, LangChain, OpenAI, PostgreSQL, Ripgrep, Git

Designed and developed an autonomous codebase research agent capable of understanding and navigating unfamiliar repositories.
Built a LangGraph-based multi-step workflow for repository cloning, indexing, planning, code exploration, evaluation, and answer synthesis.
Implemented repository mapping to create lightweight codebase summaries without loading entire projects into LLM context.
Developed an iterative search-and-evaluate loop that discovers relevant symbols, extracts code chunks, and dynamically generates new research paths.
Integrated AST-based code extraction for Python files to retrieve the smallest meaningful code regions around search matches.
Implemented repository caching and metadata persistence using PostgreSQL to avoid redundant cloning and indexing operations.
Generated evidence-based technical explanations with file and line-level citations to help engineers understand complex code flows.

GitHub Demo

AskMyPDF

Full-stack RAG application that lets users upload a PDF and hold a natural-language conversation with it, with answers grounded strictly in the document and cited by page.

Tech Used:Python, Django, Django REST Framework, LangChain, OpenAI, pgvector, PostgreSQL, HuggingFace Transformers, Docker

Designed and built a full-stack RAG pipeline that extracts, chunks, and embeds PDF content into a vector database for natural-language Q&A.
Implemented a smart ingestion engine that hashes chunks via SHA-256, so re-uploaded PDFs skip re-embedding unchanged sections and reuse stored vectors, cutting API costs and latency.
Built conversational query rewriting to resolve follow-up questions into standalone retrieval queries, improving vector search accuracy on multi-turn chats.
Integrated cross-encoder reranking (HuggingFace ms-marco-MiniLM-L-6-v2) to filter and reorder retrieved chunks by true relevance before generation.
Added a token-budget guard that trims chat history dynamically to keep prompts within the LLM's context window while preserving retrieved context.
Implemented inline, page-level citations so every answer traces back to the exact PDF page it was drawn from.
Exposed all chat mutations through a Django REST Framework API layer, with real-time streaming responses and markdown-rendered answers in the UI.

GitHub

My Skills

I am skilled in backend, frontend, and AI/LLM engineering, backed by AWS-certified cloud infrastructure experience

Programming Languages

Python, JavaScript, TypeScript, SQL

Frameworks & Libraries

Django, Django REST Framework, Celery, ReactJS, NextJS, Tailwind CSS

AI & LLM

LangChain, LangGraph, OpenAI & Anthropic APIs, RAG Pipelines, AI Agents, Prompt Engineering, Function Calling

AI Assisted Development

Claude Code

Cloud

AWS — AWS Solutions Architect – Associate Certified, Docker

Databases

PostgreSQL (pgvector), DynamoDB

Tools & Other

Git

My Certificates

I have obtained multiple certificates in technical skills, professional development, and industry-specific areas, showcasing my dedication to continuous learning and expertise.