GARIMA ✧00:00:00 UTC

Intelligent systems
built for the real world.

I'm Garima — an engineer building production-ready intelligent systems with LLMs, retrieval-augmented generation, and multi-agent architectures to transform complex workflows into reliable applications.

View Work ↓Contact ↗
garimakumari265@gmail.comLinkedIn ↗
LLMs & RAG
RETRIEVAL
MULTI-AGENT
ARCHITECTURE
PRODUCTION
SYSTEMS
↓
01 — Projects

Systems in production

Production-grade AI systems where the challenge wasn't just the model — it was the architecture, the reliability, and the intelligence at scale.

AI Research Intelligence Platform demo
01

AI Research Intelligence Platform

Production AI Platform for Scientific Literature, Adaptive Retrieval & Research Discovery

A production-grade AI research intelligence platform that transforms scientific literature into structured, searchable, and evidence-grounded research intelligence. The system combines paper ingestion, metadata enrichment, dense vector search, BM25 keyword retrieval, hybrid ranking, metadata filtering, reranking, adaptive RAG, evidence-aware reasoning, research planning, cross-paper synthesis, knowledge graphs, research gap detection, and report generation. A FastAPI backend powers the AI and retrieval infrastructure, while a Next.js, React, and Tailwind CSS frontend provides the interactive research workspace.

  • Scientific paper ingestion, normalization, chunking, metadata enrichment, and indexing
  • Hybrid retrieval combining dense vector search, BM25 keyword retrieval, metadata filtering, Reciprocal Rank Fusion, and reranking
  • Adaptive RAG pipeline that dynamically selects retrieval strategies based on query complexity and research intent
  • Evidence-grounded reasoning with source attribution, supporting and conflicting evidence, confidence signals, and citations
  • Multi-stage research orchestration covering planning, retrieval, synthesis, analysis, and structured report generation
  • Research frontier intelligence for identifying literature trends, research gaps, unresolved problems, and future directions
  • Knowledge graph capabilities for connecting papers, topics, concepts, and research relationships
  • Production observability with structured logging, request tracing, retrieval metrics, latency tracking, and AI pipeline monitoring
  • Dockerized backend and frontend architecture designed for production deployment
PythonFastAPINext.jsReactTypeScriptTailwind CSSLangGraphAdaptive RAGHybrid RetrievalLLMsPostgreSQLFAISSBM25RedisDocker
Live DemoGitHub
Multi-Agent AI for Equity Research demo
02

Multi-Agent AI for Equity Research

Enterprise AI Research Operating System for Institutional Investment

A production-grade multi-agent AI research operating system designed to automate complex equity research workflows. Orion coordinates specialized AI analysts across company research, financial analysis, industry research, news, macroeconomics, valuation, risk, evidence, and investment review. A planner-driven DAG execution engine decomposes complex research questions into coordinated tasks, while retrieval, financial data providers, MCP-enabled tools, evidence validation, critic analysis, and observability are integrated into a unified research workflow. The platform combines a FastAPI backend with a Next.js, React, and Tailwind CSS research workspace.

  • 12 specialized AI analysts covering company, financial, industry, news, macro, valuation, risk, evidence, critic, and investment research
  • Planner-driven DAG execution engine for decomposing complex research questions into coordinated agent tasks
  • Hybrid retrieval across SEC filings, financial information, market data, earnings information, and news
  • Evidence-grounded analysis with source attribution, confidence signals, structured claims, and validation
  • Dedicated critic and investment committee layers for reviewing and challenging generated research
  • MCP-enabled tool integration with a centralized tool router for external research and financial-data capabilities
  • Modular agent architecture with reusable base agents, planning templates, execution services, sandbox execution, and research pipelines
  • AI observability covering execution traces, token consumption, latency, execution costs, confidence, and agent workflow behavior
  • Production infrastructure using FastAPI, PostgreSQL, Redis, asynchronous execution, Docker, and a modern Next.js frontend
PythonFastAPINext.jsReactTypeScriptTailwind CSSMulti-Agent AILangGraphMCPRAGPostgreSQLRedisDockerFinancial AI
Live DemoGitHub
Enterprise Document Intelligence demo
03

Enterprise Document Intelligence

Production AI Platform for Intelligent Document Processing, Knowledge Retrieval & Workflow Automation

A production-grade enterprise AI platform that transforms unstructured business documents into structured, searchable, and actionable knowledge. The platform combines document ingestion, OCR, layout-aware document understanding, table and key-value extraction, semantic retrieval, conversational document intelligence, citation-backed answers, and automated AI workflows. All intelligent capabilities execute through a reusable internal AI Workflow Orchestration Engine, separating customer-facing applications from reusable AI infrastructure. The platform combines a FastAPI backend, PostgreSQL, Redis, MinIO, Docker, and a Next.js, React, and Tailwind CSS enterprise dashboard.

  • Layout-aware document understanding with OCR, parsing, table extraction, key-value extraction, and structured information extraction
  • Reusable internal AI Workflow Orchestration Engine powering document processing and intelligent automation workflows
  • Hybrid semantic retrieval for searching enterprise knowledge and generating context-aware answers
  • Conversational AI assistant with document-level evidence and citation-backed responses
  • Multi-LLM architecture supporting flexible model selection across document intelligence and workflow tasks
  • Workflow automation with visual workflow construction, approval steps, retry policies, execution states, and versioning
  • Enterprise storage and infrastructure using PostgreSQL, Redis, MinIO, FastAPI, asynchronous processing, and Docker
  • Next.js and Tailwind CSS dashboard for document processing, knowledge retrieval, workflow management, and AI-assisted operations
  • Modular architecture separating reusable AI infrastructure from customer-facing enterprise capabilities
PythonFastAPINext.jsReactTypeScriptTailwind CSSPostgreSQLRedisMinIORAGOCRMulti-LLMWorkflow EngineDocker
Live DemoGitHub
04

RLHF Self-Improving Q&A

Reinforcement Learning from Human Feedback for Adaptive QA Systems

A reinforcement learning research project exploring how human preference signals can be used to improve question-answering behavior. The system implements a feedback-driven training pipeline where model responses are evaluated through preference data, a reward model learns those preferences, and policy optimization is used to iteratively improve response quality.

  • RLHF training loop with human preference collection interface
  • Reward model trained on comparative preference data
  • PPO-based policy optimization for response quality
  • Iterative self-improvement through feedback integration
  • Evaluation framework tracking preference alignment over training rounds
PythonPyTorchRLHFPPOReward ModelingNLPTransformers
GitHub
02 — Tech Stack

Tools of the trade

The technologies I reach for when building intelligent systems that have to be fast, reliable, and observable in production.

Languages

PythonTypeScriptJavaScriptSQLHTMLCSS

AI & Machine Learning

PyTorchLLM EngineeringPrompt EngineeringContext EngineeringRAGMulti-Agent SystemsNLPReinforcement LearningAI Agents

AI Frameworks

LangGraphPydanticAIOpenAI Agents SDK

Frontend

ReactNext.jsTailwind CSS

Backend

FastAPIREST APIsWebSocketsPydanticPrisma

Databases

PostgreSQLSupabaseRedisFirebaseSQLite

Vector Databases

PineconeQdrantChromaWeaviate

DevOps & Cloud

DockerVercelRailwayRenderFly.io

Developer Tools

GitGitHubPlaywrightFirecrawlMCPn8n

Architecture & Systems

Agentic AIAI InfrastructureKnowledge SystemsMemory ArchitectureAdaptive RetrievalContext PipelinesSystem DesignAPI Design
03 — Experience

Where I've built

Professional experience shipping AI features and full-stack applications — each role a step deeper into intelligent systems.

AI/ML Engineer Intern

@ITJOBXSMay 2026 – July 2026

Developed AI-powered features to improve a platform that helps students prepare for IT careers. Worked with LLMs, APIs, Retrieval-Augmented Generation (RAG), and machine learning to enhance system performance, automate workflows, and build intelligent AI solutions.

PythonLLMsRAGMachine LearningREST APIs

Freelance Full-Stack Developer

@IndependentNovember 2024 – June 2025

Built and delivered full-stack web applications from concept to deployment, handling frontend, backend, databases, and UI design. Worked directly with clients to develop scalable, responsive, and production-ready solutions.

ReactNext.jsNode.jsPostgreSQLPrismaSQLFirebaseTailwind CSSTypeScript
04 — Writing

Notes from the margins

A mix of engineering reflection and older writing — the academic record, the published poetry, and the through-line between them.

Academics

On curiosity and the 96% in 10th grade

Reflection · 4 min

I scored 96% in my 10th and 92% in my 12th — numbers that opened doors. They led me to Miranda House, where I study Economics and Computer Science — when I joined, it was NIRF Rank 1; today it sits at NIRF Rank 2. The real story was never the marks, though. It was the quiet habit of asking one more question than the syllabus required, and an obsession with understanding human behaviour in markets — and then teaching the computer to see the same patterns I do.

Writing

Poetry in Poet's Choice — and a Jewish magazine

Published work · 5 min

Before I wrote code, I wrote poems. My prose and poetry appeared in Poet's Choice and an internal Jewish magazine — a reminder that structure, rhythm, and meaning cross every medium.

Craft

Why I write about AI the way I do

Essay · 6 min

Engineering essays and poetry share more than people think: both demand precision, both reward restraint, and both are ultimately about making complex things feel human.

05 — Contact

Let's build something
worth the orbit.

Open to AI/ML engineering roles, research opportunities, and ambitious projects. I read every message.

📧 Email💼 LinkedIn𝕏 X💻 GitHub

📩 garimakumari265@gmail.com