Jatangi Thanuja - LLM & Agents

Overview

Backend-focused engineer building end-to-end REST APIs and lightweight ML-backed document ingestion pipelines. The strongest proven skill is integrating a simple RAG pipeline with Sentence Transformers and ChromaDB as shown by the embedding generation and upsert flow in the Python ingestion code. There is limited evidence of production hardening such as robust error handling, migration histories, observability, or advanced scalability design.

Phone

Timeline

Software Development Engineer - AI Benchmarking · Middle

DevRev · Full-Time

Sep 2025 to Sep 2026 1 Year Bengaluru In office

Worked on an enterprise AI agent benchmark with tasks run in Docker sandboxes and scored by an LLM judge. Built mock MCP tools for Salesforce, Jira, Confluence, and a Zendesk-style support server, then compared Copilot, Agentforce, Claude Code, and GPT model runs using pass@10. Analyzed agent traces to diagnose failure causes and implemented reusable skills to summarize reliability results. Added benchmark data integrity checks and CI workflows for pull requests that modify benchmark datasets.

Claude Code

Copilot

Agentforce

Model Context Protocol

Python

Docker

GitHub Actions

Software Engineer Trainee - Frontend Training · Junior

MountBlue Technologies · Full-Time

May 2025 to Jul 2025 2 Months Bengaluru In office

Software Engineer - Test Automation · Middle

Rexera · Full-Time

Jul 2024 to Feb 2025 7 Months Bengaluru In office

Developed end-to-end Playwright test automation in Python for critical web application workflows. Built a weekly regression suite to validate functionality after each deployment, reducing manual QA effort. Maintained test automation to support reliable release verification.

Playwright

Python

Software Engineer Trainee - Testing Training · Junior

MountBlue Technologies · Full-Time

Apr 2024 to Jul 2024 3 Months Bengaluru In office

Projects

LLM Fundamentals in PyTorch

Implemented core components of a GPT-style language model, including tokenization, embedding layers, self-attention (Q/K/V), and a next-token training loop. Used PyTorch to set up the training process for learning token sequences. Focused on building the model architecture and training mechanics from scratch.

Enterprise HR Policy Assistant using RAG

Built a document ingestion pipeline that extracts HR policy text from PDFs and splits it into chunks while preserving source and page metadata. Created semantic embeddings with a Sentence Transformer model and stored text, embeddings, and metadata in a persistent vector database for retrieval. Indexed a subset of chunks and planned a next step for question answering with cited sources and an evaluation set.

Technical skills

Languages

4

Python

Node JS

SQL

JavaScript

Node JS

5

Express

Mongoose

Axios

Bcrypt

Dotenv

DevOps

5

GitHub Actions

Docker

GitHub

Git

Rest API

AI/ML

9

Agentforce

Claude Code

Copilot

Model Context Protocol

LLM

Claude

PyTorch

Sentence-Transformers

Transformers

Frontend

4

React.js

Redux Toolkit

Material UI

React Router

Other

9

Playwright

AI Agents

RAG

Embeddings

Harbor

Function Calling

Hallucination

Self-Attention

Tokenization

Middle Backend Developer Confidence: Medium API Engineer

Backend-focused engineer building end-to-end REST APIs and lightweight ML-backed document ingestion pipelines. The strongest proven skill is integrating a simple RAG pipeline with Sentence Transformers and ChromaDB as shown by the embedding generation and upsert flow in the Python ingestion code. There is limited evidence of production hardening such as robust error handling, migration histories, observability, or advanced scalability design.

API Design

3/10

How well APIs are designed

Basic REST API patterns and client usage are present, but there is no API versioning, inconsistent error contracts, and no idempotency or pagination strategies.

Evidence

Movie-Booking/backend/controllers/user-controller.js

Movie-Booking/frontend/src/api-helpers/api-helpers.js

Movie-Booking/backend/app.js

Data Layer & Database

3/10

Working with databases

Uses Mongoose for application data and ChromaDB for vector storage with simple upserts; lacks migration history, explicit transaction boundaries, or tuned queries/indexes.

Evidence

Movie-Booking/backend/models/User.js

enterprise-hr-policy-assistant/database/chroma_store.py

enterprise-hr-policy-assistant/ingestion/index_doc.py

Scalability & Performance

2/10

Handling load and speed

Some work on batch embedding generation and persistent vector store is present but there is little evidence of caching, queueing, connection pooling or load-testing.

Evidence

enterprise-hr-policy-assistant/retrieval/embedding.py

enterprise-hr-policy-assistant/database/chroma_store.py

System Architecture

3/10

Overall system structure

Reasonable module separation (ingestion, retrieval, DB, frontend/backend) for small projects, but no multi-service contracts, graceful degradation, or sophisticated config/secret management.

Evidence

enterprise-hr-policy-assistant/ingestion/index_doc.py

Movie-Booking/backend/app.js

Security & Auth

3/10

Protecting data and access

Basic authentication is implemented with bcrypt and JSON Web Tokens and the client sends Authorization headers, but input validation has bugs, and token lifecycle, revocation and stronger boundary validation are not implemented.

Evidence

Movie-Booking/backend/controllers/user-controller.js

Movie-Booking/backend/package.json

Movie-Booking/frontend/src/api-helpers/api-helpers.js

Reliability & Observability

2/10

Stability and monitoring

Minimal reliability and observability practices; try/catch blocks and console logs exist but structured logging, timeouts, retries with backoff, and shutdown handling are absent.

Evidence

Movie-Booking/backend/controllers/user-controller.js

enterprise-hr-policy-assistant/ingestion/index_doc.py

Verified artifacts

sasikalanalagatla/3_MB· PR

Expertise

Backend AI & LLM· Middle

Databases & Vector Storage· Middle

Node.js· Middle

Industries

Media & Entertainment· Middle

Technologies

Python· since 2024 · Middle

Rest API

Dotenv

Recommendations

Owner of small-to-medium REST API services and CRUD endpoints including user, booking and admin flows.
Prototype document search and RAG features that integrate embedding generation and vector stores for enterprise knowledge use cases.
Harden services for production by adding structured logging, input validation fixes, migration tooling, and resilience patterns such as timeouts and retries.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

enterprise-hr-policy-assistant Movie-Booking

Junior Frontend Developer Confidence: Medium Fullstack

Fullstack web developer (Junior) focused on React and Express with practical end-to-end movie booking features. The strongest proven skill is building interactive React UIs wired to an API layer, evidenced by frontend/src/components/Header/Header.js and frontend/src/api-helpers/api-helpers.js. There is limited evidence of advanced state discipline, automated tests, measured performance optimization or deliberate accessibility work.

UI Component Architecture

3/10

How interface parts are built

Basic component decomposition using React and MUI with conventional props and local state; no bespoke design system or advanced composition patterns.

Evidence

frontend/src/components/Header/Header.js: Header component using MUI Tabs, Autocomplete and react-router links

frontend/src/components/Movies/AddMovie.js: AddMovie form component with local state and actor list handling

frontend/src/profile/UserProfile.js: UserProfile using MUI List to render bookings and API-driven state

Responsive & Cross-browser

2/10

Works on all screens and browsers

Responsive layout is implemented via MUI Box and percent widths but lacks advanced responsive strategies or explicit RTL/i18n and feature-detection handling.

Evidence

frontend/src/components/Movies/AddMovie.js: form container uses Box width 50% and flex layout

frontend/src/profile/UserProfile.js: profile layout splits 30%/70% with MUI Box and flexDirection

Performance Optimization

1/10

Speed of the interface

No measurable/performance-driven work; no request cancellation, virtualization, code-splitting or bundle analysis present.

Evidence

frontend/src/api-helpers/api-helpers.js: central axios helpers with no cancellation, caching or performance concerns handled

Accessibility & Semantics

1/10

Usable for everyone

Uses accessible MUI primitives but no bespoke accessibility enhancements, ARIA on custom widgets, or focus/keyboard management for custom controls.

Evidence

frontend/src/components/Header/Header.js: Autocomplete + TextField usage (library-a11y inherited, no custom a11y handling)

frontend/src/profile/UserProfile.js: List of bookings rendered with MUI components but no explicit focus management

State Management & Data Flow

2/10

Managing data in the app

Simple client-server state flows with axios helpers and react-redux referenced; lacks sophisticated server-state discipline like cancellation, optimistic updates, or state machines.

Evidence

frontend/src/api-helpers/api-helpers.js: axios wrappers for CRUD and auth endpoints

frontend/src/components/Header/Header.js: useEffect fetching getAllMovies and navigation on selection

backend/controllers/user-controller.js: express controller endpoints implementing user CRUD and bookings retrieval

UX & Visual Polish

3/10

Look and feel quality

Reasonable visual polish using MUI and icons, usable forms and lists, but missing skeletons, explicit loading/empty states and undo patterns.

Evidence

frontend/src/components/Movies/AddMovie.js: styled form with MUI Typography, Box, and Button

frontend/src/profile/UserProfile.js: bookings list with icons and styled ListItem

frontend/src/components/Header/Header.js: search Autocomplete and app-level navigation with Tabs

Expertise

React· Junior

Industries

Media & Entertainment· Middle

Technologies

Node JS· Middle

Express

Bcrypt

React.js

Mongoose

Axios

Material UI

Redux Toolkit

React Router

JavaScript· mentioned only

Recommendations

Develop small-to-medium fullstack CRUD applications using React, Express and Mongoose such as booking or admin dashboards.
Own feature work that includes client-side state discipline - add request cancellation, loading/empty/error states and optimistic updates with rollback.
Improve engineering quality by adding unit and integration tests, explicit input validation, and stronger error handling on both client and server.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Movie-Booking

Job preferences

Work permits: India