SrinathMuppala
- Agentic AI Systems
- RAG & LLM Fine-Tuning
- Production Backends
- Cloud Infrastructure
- ML / DL / CV / NLP
I’m the kind of engineer who sees someone doing the same boring thing twice and immediately thinks, “There has to be a way to automate this.” Basically, I like making computers do the boring stuff so humans can work on the interesting stuff.
I love building AI systems that actually do something—not just another chatbot wrapped around an API. I’ve built everything from an AI platform that turns architectural floor plans from a two-day manual process into minutes, to GymBuddy, a voice-first AI workout companion I built because I was tired of manually logging workouts at the gym.
What excites me most is taking a messy real-world problem, figuring out where AI actually helps, and then owning the whole thing: agents, models, APIs, databases, infrastructure, deployment, and the user experience.
Experience
Where I've Built
AI Engineer
InovonicsBoulder, COSept 2025 – Apr 2026
AI-assisted floor plan vectorization
- Reduced the time taken to map and plot an architectural floor plan from 2 days to a matter of minutes.
- Developed a full-stack AI-powered floor plan vectorization platform using YOLOv12-OBB for structure detection, with a FastAPI backend, Celery workers, a Redis task queue, a PostgreSQL database, and a Next.js frontend deployed on GCP.
- Overcame small-object detection limits in high-resolution engineering schematics with tiled sliding-window inference and overlap-based deduplication, plus a 9-stage OpenCV preprocessing pipeline that cleans and normalizes noisy architectural drawings before ML inference.
- Designed Shapely-based room boundary extraction (wall polygon union, Minkowski gap closing, free-space set-difference) and optimized receiver placement as a set-cover problem using OR-Tools CP-SAT with 180-ray coverage analysis.
- YOLOv12-OBB
- FastAPI
- Celery
- Redis
- PostgreSQL
- Next.js
- OpenCV
- Shapely
- OR-Tools CP-SAT
- Google Cloud
Education
University of Colorado Boulder
Master of Science in Computer Science — Artificial Intelligence
Aug 2024 – May 2026GPA 3.87 / 4.0
SRM Institute of Science and Technology
Bachelor of Technology in Computer Science and Engineering
Sept 2020 – May 2024GPA 3.98 / 4.0
Featured Work
Selected Builds
Twelve builds, dealt as a hand — hover a card to turn it over, click it to read the hand.
GymBuddy
Voice-first AI workout companion
PricePilot
LLM fine-tuning and high-throughput serving
DealHunter AI
Multi-agent deal discovery engine
Document Copilot
Financial document RAG platform
GraphFusion RAG
Hybrid graph + vector retrieval
Production LLM API
FastAPI + LangGraph endpoint
Autonomous Traders
Multi-agent AI stock trading system
BuffRelay
Distributed WebSocket messaging
Music-Separation K8s
Kubernetes audio stem pipeline
ARGONAUT
Vision-language QA on VizWiz
Clarity
Unmasking political question evasions
CourtVision
Player, ball and court tracking
Skillset
The Stack
Nine clusters, drifting — hover one to read the techniques behind it.
- Python
- TypeScript
- JavaScript
- C++
- Java
- Bash
- HTML
- CSS
- SQL
- OpenAI
- MCP
- LangChain
- LangGraph
- Hugging Face
- vLLM
- Modal
- Meta LLaMA 3.2
- Pydantic
- Agentic AI
- OpenAI Agents SDK
- function & tool calling
- multi-agent orchestration
- intent classification & routing
- LiteLLM
- streaming responses
- prompt caching
- structured outputs
- GPT-4o Vision
- prompt-injection detection
- PII redaction
- LoRA / PEFT
- QLoRA (4-bit NF4)
- bitsandbytes
- TRL / SFTTrainer
- pgvector
- ChromaDB
- OpenAI Embeddings
- hybrid retrieval
- Postgres full-text search
- Reciprocal Rank Fusion
- LLM reranking
- query rewriting
- knowledge-graph traversal / GraphRAG
- multi-hop Q&A
- hierarchical chunking
- idempotent batch embedding upserts
- citation verification & self-correction
- text-embedding-3-large
- PyTorch
- TensorFlow
- scikit-learn
- NumPy
- pandas
- OpenCV
- CUDA
- Transformers
- YOLO (v5x / v8 / v12, OBB)
- ResNet-50
- ViT
- BERT
- DeBERTa-v3
- LSTM decoders
- attention fusion
- multi-task learning
- warm-start / transfer learning
- tiled sliding-window inference
- homography-based spatial filtering
- Shapely geometry
- OR-Tools CP-SAT
- FastAPI
- Django
- Flask
- Node.js
- Express.js
- Spring Boot
- Socket.IO
- Celery
- Redis Streams
- Kafka
- Swagger
- OpenAPI
- Pydantic
- SQLAlchemy
- Prisma ORM
- pytest
- OAuth2 / OIDC
- JWT & JWKS
- REST
- Alembic
- asyncio
- BCrypt
- keyset pagination
- connection pooling
- response caching with TTL
- outbox pattern
- PostgreSQL
- MySQL
- MongoDB
- Redis
- Upstash Redis
- Supabase
- MinIO
- ChromaDB
- pgvector
- Supabase Auth & RLS
- Cloud SQL
- Cloud Redis
- Google Cloud
- AWS
- Docker
- Kubernetes
- Terraform
- GitHub Actions
- Railway
- Modal
- Git
- Apache Maven
- Cloud Run
- App Engine
- Compute Engine
- ALB
- CloudFront
- CloudWatch
- Weights & Biases
- MLflow
- k6 load testing
- LangSmith
- LangSmith tracing
- LLM-as-a-judge pipelines
- MRR / nDCG / keyword coverage
- structured JSON logging
- latency, error-rate & token-usage metrics
- CloudWatch
- UptimeRobot
- model checkpointing
- Next.js
- React
- TypeScript
- Tailwind CSS
- Gradio
- Web Speech API (wake word, TTS)
9 clusters
73 tools across the stack — hover a cluster for the techniques behind it.