Suraj Biswas — LLM · Agentic AI

// about

cat about.txt

about.txt ~/.config/suraj/

# summary

Software developer specializing in backend systems, GenAI, and agentic AI — building scalable, secure platforms with production LLM pipelines.

# what i do

Agentic RAG & multi-hop search · LangGraph prompt optimization · RAG deviation analysis · FastAPI backends with Bedrock, LiteLLM & Langfuse observability.

# background

4+ years at Infosys & Cognizant — from Django/PostgreSQL backends to owning enterprise GenAI modules end-to-end. B.Tech CSE, MAKAUT · GPA 8.7 · Kolkata.

// experience

git log experience/

Jan 2026 — Present Bhubaneswar

Sr. Associate Consultant · Infosys Ltd.

Agentic AI Agentic RAG LLM FastAPI

Enterprise GenAI Contract Analysis Platform

Built an agentic multi-hop Deep Search engine that autonomously generates sub-queries to close knowledge gaps, consolidates findings against configurable risk registers, and produces grounded counter-proposals from SME guidance documents.
Designed two Agentic AI prompt-improvement systems: a LangGraph pipeline with LLM-as-judge for real-time critique/refinement, and a RAGAS-powered optimization framework tuned on SME-validated ground truth (TPR/FPR 85%/25%) using Claude Sonnet via AWS Bedrock through LiteLLM.
Owned a RAG-based deviation analysis module — vector retrieval, Jaccard similarity scoring, and LLM reasoning to flag missing obligations and compliance gaps with AI-recommended redlines (~30% less manual clause comparison).
Architected plugin-based workflow orchestration for 4 GenAI capabilities (document upload, Q&A, risk analysis, deviation analysis) with priority-ordered handlers, feature-level failure isolation, and real-time SSE progress.
Shipped 20+ REST endpoints across 5 API routers with FastAPI, SQLAlchemy 2.0, and PostgreSQL RLS for multi-tenant isolation; enforced 85% test coverage and monitored LLM latency, cost, and traces in production with Langfuse.

Sep 2021 — Jan 2026 Bangalore

Jr. Software Engineer · Cognizant Technology Solutions

Django PostgreSQL React

Financial Report Visualization & Repository Platform

Improved front-end performance by 30% and user engagement by 25% using React.js, Redux, and optimized component rendering.
Optimized Django and PostgreSQL queries with stored procedures, reducing data retrieval time by ~30% for large financial datasets.
Built RESTful APIs with Django REST Framework, achieving ~20% faster request-response cycles and smoother downstream integration.

// projects

ls projects/

in development

StoryGen

github ↗

AI-powered software planning platform that transforms BRDs into features, sub-features, user stories, story points, and test cases through a structured agent-based pipeline with human review at every stage.

Agent Orchestration RAG FastAPI LiteLLM PostgreSQL React

open source

multi-hop-research-engine

github ↗

Multi-hop RAG chatbot with iterative query refinement across PDFs — sub-second retrieval over 10,000+ chunks with 28–35% better accuracy than single-hop. Streamlit UI cut manual research time by ~40–60% for domain experts.

Agentic RAG LangChain FAISS Gemini Python

open source

Image-Captioning-Bot

github ↗

Encoder-decoder pipeline generating natural-language captions from images using a ResNet50-based deep learning model.

Python TensorFlow ResNet50

→ more on github.com/sb-robo

// skills

cat skills.json

llm & agentic ai

Agentic AI Agentic RAG LangGraph LangChain AWS Bedrock LiteLLM RAGAS LLM-as-judge

rag & retrieval

Multi-hop RAG VectorDB FAISS Embeddings Semantic Search

backend & databases

FastAPI Django Django REST PostgreSQL SQLAlchemy 2.0

cloud & observability

AWS Docker GitHub Actions Pytest Langfuse OpenTelemetry

languages

Python JavaScript

// contact

mail -s "hello"

Open to roles and conversations around LLM systems, agentic RAG, and production GenAI backends.

email me resume.pdf github linkedin