Pranavsai Padala - AI Engineer

Overview

A pragmatic LLM-focused engineer (mid-level) who builds applied agent pipelines and backend services for data and document processing. The strongest proven skill is building LLM-backed data ingestion and query tooling, demonstrated by the FastAPI NL2SQL backend (nl2sql/backend/main.py) that maps uploaded CSV/Excel into SQLite, detects relationships, prompts an LLM for SQL, executes it and returns structured results. There is limited evidence of model training, experiment tracking, production MLOps hardening, or performance optimization in public code.

Phone

Timeline

Jawaharlal Nehru Technological University Hyderabad (JNTUH)

Bachelor's Degree · Computer Science and Engineering

2022–2026 Hyderabad, Telangana

Programmer Analyst Trainee · Junior

Cognizant Technology Solutions · Full-Time

Jan 2026 to May 2026 4 Months Chennai In office

Built scalable web applications and REST APIs using C#/.NET and ADO.NET to address business needs. Worked with Azure Functions, Logic Apps, Storage Accounts, and API Management, using Application Insights for monitoring. Applied Azure security practices using Key Vault and Managed Identity to protect secrets and control access.

C#

Rest API

Azure

Projects

EduMind: AI-Powered Study Companion

Built a web-based AI study companion using React and FastAPI to provide explanations, quizzes, and note summarization. Used a locally hosted LLM (Mistral via Ollama) to generate study content. Implemented privacy-focused retrieval using RAG with FAISS/ChromaDB-style vector search over study materials.

NL2SQL: AI Assistant for Natural Language Querying

Designed and built an LLM-based NL2SQL system that converts natural language requests into SQL and runs them against structured datasets like CSV and Excel. Implemented a full-stack application with an interactive UI and an LLM-driven backend for query processing. Added visualization of query results to help users interpret outputs.

Azure Retail Inventory Management System

Developed a cloud-based inventory system using an ASP.NET Core Web API deployed to an Azure Web App. Used Azure API Management for secure routing and authentication. Integrated Service Bus, Logic Apps, and Azure Key Vault with Managed Identity to automate low-stock email alerts and protect secrets, storing data in Azure Table Storage.

Technical skills

Languages

4

Python

Java

SQL

C#

AI/ML

10

LLM

Transformers

Pandas

Google GenAI SDK

LangChain

LangGraph

Mistral

Hugging Face

NumPy

Scikit-learn

Frontend

3

React.js

Tailwind CSS

Three.JS

DevOps

4

Azure

Netlify

Vercel

Rest API

Other

14

FastAPI

ASP.NET Core

SQLite

FAISS

Ollama

GitHub

AI Agents

Git

Deep Learning

Machine Learning

RAG

Embeddings

Prompt Engineering

NLP

Middle AI/ML Engineer Confidence: Medium LLM Engineer

A pragmatic LLM-focused engineer (mid-level) who builds applied agent pipelines and backend services for data and document processing. The strongest proven skill is building LLM-backed data ingestion and query tooling, demonstrated by the FastAPI NL2SQL backend (nl2sql/backend/main.py) that maps uploaded CSV/Excel into SQLite, detects relationships, prompts an LLM for SQL, executes it and returns structured results. There is limited evidence of model training, experiment tracking, production MLOps hardening, or performance optimization in public code.

Model Architecture & Training

1/10

How well models are designed and trained

No custom model architectures or training loops; only uses pretrained transformers pipelines for classification and sentiment with no training code or hyperparameter design.

Evidence

multi-agent-ai-system/agents/classifier_agent.py: pipeline('text-classification', model='distilbert-base-uncased-finetuned-sst-2-english')

multi-agent-ai-system/agents/email_agent.py: pipeline('text-classification', model='distilbert-base-uncased-finetuned-sst-2-english')

Data Pipeline & Feature Engineering

4/10

How data is prepared for models

Practical data ingestion and schema mapping for CSV/Excel, heuristic relationship detection across tables, and document text extraction and field parsing; solid applied data plumbing but no advanced feature engineering.

Evidence

nl2sql/backend/main.py: generate_create_table_sql(df: pd.DataFrame, table_name: str)

nl2sql/backend/main.py: upload_csv endpoint reading CSV/XLSX, creating in-memory sqlite and df.to_sql

nl2sql/backend/main.py: find_relationships(dfs: Dict[str, pd.DataFrame])

Experimentation & Evaluation

1/10

How results are measured and tested

No experiment tracking, metrics, validation pipelines, or systematic evaluation artifacts present; only minimal test stubs/imports.

Evidence

multi-agent-ai-system/test_main.py: tests present but only import pdfplumber (no metrics or tracking)

multi-agent-ai-system/test_pdf.py: import pdfplumber (no evaluation harness)

MLOps & Deployment

3/10

How models are shipped to production

Contains deployment-ready elements like a FastAPI app with endpoints, session handling, and run command, plus simple integration with an external LLM API, but lacks production-grade concerns (auth, monitoring, model versioning, retries).

Evidence

nl2sql/backend/main.py: FastAPI endpoints (/upload, /ask), uvicorn.run at __main__

nl2sql/backend/main.py: llm_generate() with google.generativeai integration and GEMINI_API_KEY handling

Computational Efficiency

1/10

How efficiently computing resources are used

No evidence of GPU/memory optimizations, batching strategies, quantization, or profiling; standard use of transformers and torch without efficiency engineering.

Evidence

multi-agent-ai-system/requirements.txt: transformers and torch listed but no optimization code

multi-agent-ai-system/agents/*.py: use of transformers.pipeline without batching/optimization

Research Depth & Innovation

1/10

Depth of research and new ideas

No research implementations, paper repros, or novel algorithms; solutions are pragmatic and heuristic rather than research-level innovations.

Evidence

multi-agent-ai-system/agents/pdf_agent.py: simple rule-based invoice field extraction

nl2sql/backend/main.py: find_relationships heuristic and prompt-based SQL generation

Expertise

AI / LLM Engineering (Agents)· Middle

Industries

Data & Analytics· Middle

Financial Services· Middle

Technologies

Deep Learning

Python· since 2024 · Middle

SQL· since 2026 · Junior

LangGraph

LangChain

FAISS

Embeddings· since 2026

Scikit-learn· since 2026

Prompt Engineering· since 2026

AI Agents

NLP· since 2026

Google GenAI SDK

Ollama

Mistral

Transformers

Pandas· since 2026

NumPy· since 2026

Git

SQLite

LLM

RAG· since 2026

Hugging Face

GitHub

Machine Learning· since 2026

SQL· mentioned only

Recommendations

Develop prototypes of LLM-powered data tooling and extraction pipelines that integrate model calls, structured storage, and simple UIs (e.g., NL2SQL and document extraction pipelines).
Implement hardened serving patterns for LLM backends: retries/backoff, auth, input sanitization and model fallback strategies to make agent workflows production-ready.
Build evaluation and observability: add test suites, metrics, and simple experiment tracking (W&B/MLflow) for LLM prompts and extraction accuracy.
Extend document extraction with robust parsing and validation (schema checks, unit tests) for higher-assurance financial/invoice processing.

Repositories

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

Pranavsai1410 nl2sql multi-agent-ai-system

Middle Backend Developer Confidence: Medium API Engineer

Backend API engineer at a middle level specializing in building LLM-backed data tools and educational services. Proven ability to integrate LLMs into backend flows and translate user uploads into queryable data, demonstrated by magic_pipeline and llm_generate in nl2sql/backend/main.py and the ChatOllama pipeline initialization in EduMind/EduMind-backend/main.py. Not evidenced is experience with large-scale system design, production-grade security hardening (token lifecycle and SQL safety), or formal DB migration histories and load testing.

API Design

4/10

How well APIs are designed

Basic, consistent REST endpoints with reasonable error responses and cookie-based session handling, but no versioning, idempotency or pagination strategy.

Evidence

nl2sql/backend/main.py: upload_csv and ask_question endpoints with HTTPException usage and session cookie handling (get_or_create_session_id)

EduMind/EduMind-backend/main.py: /api/register and /api/login endpoints with input validation and structured JSON responses

Data Layer & Database

4/10

Working with databases

Practical use of databases (in-memory SQLite and MongoDB) and programmatic schema generation, but no transactional boundaries, explicit migrations for DB engines, or advanced query tuning.

Evidence

nl2sql/backend/main.py: generate_create_table_sql + df.to_sql into sqlite3 in-memory and pd.read_sql_query usage

EduMind/EduMind-backend/main.py: PyMongo usage for user creation/login and mongo.db.command('ping') checks

Scalability & Performance

3/10

Handling load and speed

Some performance-minded choices (upload size limits, PDF page limits, simple retry loops) but no caching/invalidation strategy, connection pooling, or queue-based decoupling.

Evidence

EduMind/EduMind-backend/main.py: app.config['MAX_CONTENT_LENGTH'] = 50MB and extract_text_from_pdf limits pages processed to 50

EduMind/EduMind-backend/main.py: retry loops in parse_plain_text_to_json and generate_quiz to attempt regenerating missing questions

System Architecture

3/10

Overall system structure

Reasonable module separation and clear LLM pipeline initialization and middleware usage, but no evidence of deliberate service decomposition, config/secrets management at scale, or multi-service contracts.

Evidence

nl2sql/backend/main.py: magic_pipeline composing schema info and relationship detection before LLM call

EduMind/EduMind-backend/main.py: initialise_model building ChatOllama pipelines and CORS plus environment variable checks for MONGO_URI/SECRET_KEY

Security & Auth

3/10

Protecting data and access

Basic auth practices present (bcrypt password hashing, JWT tokens with expiry, cookie flags), but missing refresh/revocation flows and there are risky patterns like executing LLM-produced SQL without sanitization and potential hard-coded API key placeholders.

Evidence

EduMind/EduMind-backend/main.py: register/login use bcrypt.hashpw and jwt.encode with expiration

nl2sql/backend/main.py: llm_generate + magic_pipeline execute LLM-generated SQL via pd.read_sql_query and GEMINI_API_KEY placeholder present

Reliability & Observability

4/10

Stability and monitoring

Good observability for a small project: file and console logging, health endpoint and defensive error handling across endpoints, but limited operational resilience (no timeouts, circuit breakers, or graceful shutdown patterns).

Evidence

EduMind/EduMind-backend/main.py: logging setup with FileHandler and health_check endpoint reporting pipeline and mongo availability

nl2sql/backend/main.py: try/except blocks around upload_csv and ask_question with HTTPException and printed session logs

Expertise

Backend AI & LLM· Middle

Industries

Education· Middle

Technologies

C#· since 2026 · Junior

Rest API· since 2026

FastAPI

ASP.NET Core

Stack· mentioned only

Recommendations

Build small-to-medium LLM-backed APIs and prototypes that process user data and return structured results (e.g., NL2SQL, summarization, quiz generation).
Harden authentication and token lifecycle: add refresh tokens, revocation, scoped tokens and stronger token parsing/validation patterns.
Add safeguards for executing model-generated SQL: implement SQL whitelisting, sandboxing, or a query AST validator to prevent unsafe execution.
Invest in production readiness work: containerization, health/metrics export, request timeouts, graceful shutdown and simple load-testing to validate scaling assumptions.

Repositories

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

Pranavsai1410 nl2sql EduMind

Middle Frontend Developer Confidence: High App Engineer

Frontend-focused app engineer (middle) who builds polished, interactive single-page experiences and in-browser AI demos with strong visual and performance attention. The strongest proven skill is building interactive, performance-minded frontends and in-browser RAG/LLM demos, evidenced by the custom TF-IDF retrieval, chat UI and three.js shader scene in the portfolio index.html. There is little or no evidence of automated tests, CI-quality infrastructure for backend services, or multi-repo system design in the public code provided.

UI Component Architecture

4/10

How interface parts are built

Custom DOM-driven component patterns and small reusable UI pieces, but no framework/component system or tests; composition is manual via functions like renderAll and initUI.

Evidence

Pranavsai1410/Pranavsai1410.github.io/index.html:renderAll (builds skills, timeline, projects DOM) Pranavsai1410/Pranavsai1410.github.io/index.html:initUI (cert flip, nav, typed roles, custom cursor) Pranavsai1410/Pranavsai1410.github.io/index.html:initSphere (3D skill globe DOM projection)

Responsive & Cross-browser

6/10

Works on all screens and browsers

Deliberate responsive adjustments and feature detection with graceful fallbacks (reduced-motion, non-WebGL) and media-query based layout changes.

Evidence

Pranavsai1410/Pranavsai1410.github.io/index.html:CSS responsive media queries (various @media blocks) Pranavsai1410/Pranavsai1410.github.io/index.html:initScene (checks for window.THREE and fallbacks to body.no-webgl), matchMedia use for reduced-motion

Performance Optimization

6/10

Speed of the interface

Measured performance-aware choices: rAF loop with lightweight loop array, devicePixelRatio tuning, geometry/detail reduction on small screens, and IntersectionObserver to avoid offscreen work.

Evidence

Pranavsai1410/Pranavsai1410.github.io/index.html:initScene (renderer.setPixelRatio, reduced geometry for small screens, isSmall branch for particles) Pranavsai1410/Pranavsai1410.github.io/index.html:main frame loop and loops array (requestAnimationFrame driven, with try/catch per-loop and visibility checks)

Accessibility & Semantics

6/10

Usable for everyone

Solid accessibility attention: ARIA labels, aria-live for chat, keyboard interaction for custom widgets, focus management and prefers-reduced-motion handling.

Evidence

Pranavsai1410/Pranavsai1410.github.io/index.html:msgs element has aria-live and ask form focuses input after response (initRAG.ask) Pranavsai1410/Pranavsai1410.github.io/index.html:initUI (cert flip: keydown handler for Enter/Space, accessible nav and role attributes)

State Management & Data Flow

4/10

Managing data in the app

Custom client-side state and retrieval pipeline (TF-IDF vectors, search, focus state) with basic concurrency guards but no advanced cancellation/rollback or server-state caching patterns.

Evidence

Pranavsai1410/Pranavsai1410.github.io/index.html:initRAG (tok, vec, dv, search functions implementing TF-IDF + cosine) Pranavsai1410/Pranavsai1410.github.io/index.html:ask (chat flow with busy flag, setStep pipeline, viz.focus integration)

UX & Visual Polish

6/10

Look and feel quality

High visual polish and UX touches: boot sequence, toasts, typing animation, screenshot carousel with pause-on-hover, 3D tilt and interactive globe; graceful empty/no-match flows included.

Evidence

Pranavsai1410/Pranavsai1410.github.io/index.html:boot (boot lines, progress bar) and toast (copyMail interaction) Pranavsai1410/Pranavsai1410.github.io/index.html:loadShots (screenshot carousel with pause and nav) and initUI (tilt interactions)

Expertise

Frontend AI Integration· Middle

Web Performance & Optimization· Middle

Industries

Commerce· Middle

Education· Middle

Technologies

Tailwind CSS

Vercel

Netlify

Azure· since 2026

Three.JS

React.js

Recommendations

Build interactive frontends and in-browser ML demos (RAG prototypes, demo UIs for LLMs and vector search).
Implement production-ready client-server integrations for LLM backends and Azure-hosted services (API contracts, auth, observability).
Lead small UX-focused features that require WebGL or canvas performance tuning and accessible custom widgets.
Polish engineering hygiene: add unit/integration tests, CI checks (lint, a11y, bundle analysis) and documentation for the retrieval/chat flow.

Repositories

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

Pranavsai1410.github.io EduMind EduMind

Job preferences

Work permits: India
Employment type: Full-Time
Locations: Hyderabad, Bengaluru, Chennai