Available for AI engineering roles

Surya Prakash Vankudothu

I design and ship production AI systems — LLM applications, RAG pipelines, computer vision models and autonomous agents that solve real business problems.

View Projects Download Resume Contact Me

About

Engineering intelligence into products

I turn research-grade AI into reliable, measurable systems — from data pipeline to deployed endpoint.

Based in Hyderabad , Telangana · Open to relocation & remote

LLM Applications

Production chat, extraction and reasoning systems built on OpenAI, Gemini and open-weight models.

RAG Systems

Hybrid retrieval, chunking strategy, re-ranking and evaluation over Milvus, FAISS and pgvector.

AI Agents

Stateful, tool-calling agents orchestrated with LangGraph and guarded by deterministic workflows.

Computer Vision

Detection, recognition and VLM fine-tuning with YOLO, DETR, InsightFace and Qwen2.5-VL.

Machine Learning

End-to-end modelling: feature engineering, training, evaluation and drift-aware monitoring.

Experience

Where I've shipped AI

Hands-on work across generative AI, computer vision and business automation.

AI Engineer

Ayurzen Pharma Pvt Ltd

Jan 2026 — July 2026

Developed an AI email automation agent for inbound query triage, drafting, and intelligent routing
Built an AI-powered video generation tool to automate marketing content creation workflows
Conducted extensive research and authored Detailed Project Reports (DPR) for the Pharma Hub project
Designed and optimized the pharma sales website user interface and conversion flow
Integrated Generative AI tools and internal dashboards to streamline business operations and sales

LLM Trainer

Deccan AI

Nov 2025 — Present

Trained LLMs and evaluated outputs (Image, Video, Audio annotations) using RLHF, improving response quality, instruction adherence, and accuracy
Applied rubric-based scoring to assess relevance, truthfulness, groundedness, and tool calling accessibility
Performed detailed prompt analysis, fact-checking, and tool integration testing to minimize hallucinations
Assessed complex prompts and multi-step reasoning chains, providing precise evaluations that improved overall model reliability

Machine Learning Intern

FastPix.io

June 2025 — July 2025

Fine-tuned Qwen2.5-VL for domain-specific visual understanding
Built computer vision models for production inference
Shipped a face recognition system with embedding search
Owned dataset preparation, cleaning and annotation pipelines
Ran distributed model training experiments
Designed evaluation harnesses and benchmark reporting

Skills

The toolkit behind the work

Depth where it counts — modelling, retrieval, vision and the backend that serves it all.

Programming

PythonC++SQL

Machine Learning

PyTorchTensorFlowScikit-LearnXGBoost

Generative AI

LangChainLangGraphLlamaIndexOpenAI APIHugging FaceRAGVector DatabasesPrompt Engineering

Computer Vision

OpenCVYOLODETRInsightFace

Backend

FastAPIFlaskREST APIs

Database

PostgreSQLMongoDBMilvusFAISS

Cloud & DevOps

DockerGitGitHubAWS

Featured Projects

Systems I've built end-to-end

Agentic pipelines, vision models and automation that went beyond the notebook.

Generative AI

AI-Powered Talent Scouting Agent

An intelligent recruiting agent that ingests Job Descriptions, discovers matching candidates, simulates recruiter outreach, and returns a calibrated shortlist.

FastAPIStreamlitPostgreSQL/pgvectorRRF FusionCross-EncoderGemini 2.5 FlashExplainability

GitHub Live Demo

Computer Vision

Qwen2.5-VL Multimodal Face & Audio AI

Fine-tuned Qwen2.5-VL-32B with QLoRA on L40s GPUs for Indian celebrity face recognition, Whisper-large-v3 audio transcription, and Milvus vector RAG.

Qwen2.5-VL-32BQLoRABuffalo-LWhisper-large-v3PyTorch/CUDAMilvusFastAPI

GitHub Live Demo

Analytics

Multi-Branch Sales Analytics & BI Dashboard

Full-stack retail intelligence platform across 5 branches (Hyderabad & AP) with real-time revenue, P&L profit tracking, tax audits, and salesperson leaderboards.

React 18TypeScriptNode.js/ExpressPostgreSQLRechartsReact QueryTailwind CSS

GitHub Live Demo

Computer Vision

AdaCLIP-D: Hybrid Denoising Anomaly Detection

A multi-stage hybrid denoising framework combining spatial U-Net restoration with in-ViT DnCNN modules for robust Zero-Shot Anomaly Detection under severe noise.

AdaCLIPU-Net DenoiserDnCNN ModulesPolarized Self-AttentionPyTorchMVTec AD / VisA

GitHub Live Demo

Leadership

Position of responsibilities

Roles I took on during college — from cultural fests to club coordination.

Overall Co-ordinator

Pixxel IITP

Photography and film making club of IIT Patna

Co-ordinator

Production Committee

The Annual Cultural Fest ANWESHA, IIT Patna

Co-ordinator and PoC

Events Committee

INTER IIT Culturals 7.0, IIT Patna

Co-ordinator

Flagship Events Committee

Annual tech fest CELESTA

Co-ordinator

Events Committee

Annual sports fest INFINITO

Achievements

Numbers that back the work

AI Projects Completed

0 mo

Internship Experience

Models Fine-tuned

GitHub Repositories

LeetCode Problems Solved

Tech Stack

Daily drivers

The frameworks, stores and platforms I reach for first.

Python

PyTorch

LangChain

LangGraph

FastAPI

Docker

AWS

GitHub

OpenAI

HuggingFace

MongoDB

PostgreSQL

Redis

Milvus

FAISS

Contact

Let's build something intelligent

Open to AI engineering roles, freelance builds and research collaborations.

Elsewhere

LinkedIn GitHub Email LeetCode

Hyderabad , Telangana