Devaj Mody

Devaj Mody

Backend & AI engineer building intelligent systems

bio

I'm a backend and AI engineer with 5+ years building production systems. Graduating April 2026 with my MS in Software Engineering from RIT — currently training LLMs, building multi-agent systems, and working on computer vision for robotics. Previously led engineering at PhyFarm (AI-powered farm automation across 500+ farms), built ML infrastructure at Purplle (7M+ MAUs), and scaled AdTech at BookMyShow (5M+ MAUs).

stack

Agentic Coding Claude Code · Cursor
AI & ML LangGraph · PyTorch · Unsloth · RAG · Pinecone · Claude API
Languages Python · Go · JavaScript
Infrastructure Kubernetes · Docker · AWS · GCP · Kafka · Terraform
Data BigQuery · Bigtable · InfluxDB · PostgreSQL · MongoDB · Redis
Hardware PLC · Modbus/RS485 · I2C/SPI · MQTT · ROS
2026
WF

Building an image generation automation pipeline at Wallflower Merchandise. Fine-tuning Flux models using LoRA and QLoRA for targeted product image modifications. Working with vision-language models and CLIP for multi-modal understanding.

LoRA QLoRA Flux CLIP VLMs Python
2024 -

Pursuing my MS in Software Engineering at Rochester Institute of Technology. Coursework focuses on training and fine-tuning LLMs, neural network architectures, interpretability, VLAs, and agent AI.

Glaze Chemistry LLM — Trained Qwen3-4B with GRPO and SFT using Unsloth on 6,153 ceramic glaze recipes. Iterated reward function through 4 versions, preventing reward hacking with synonym resolution and 100% validation gating. Built chemistry-based evaluation automating reward signals without human labels.
Unsloth Qwen GRPO PyTorch RL
Endo Automation — AI-powered farm startup. LangGraph multi-agent system with RAG pipeline indexing 50K+ docs. Deployed on AWS/Kubernetes with 99.9% uptime. Won Sarah Ramsey Strong Fund recognition.
LangGraph Pinecone Go Kubernetes
2022 - 2024

Led a 5-engineer team at PhyFarm shipping AI-powered farm automation to 500+ farms. Built RAG pipeline on 100K agricultural documents for crop diagnostics serving 5K+ queries/month.

RL Nutrient Controller — Custom actor-critic RL agent for nutrient dosing, achieving target EC/pH 60% faster than PID controllers, saving $2K per installation.
PyTorch RL Embedded
IoT & Embedded Platform — Event-driven system with Go & MQTT handling 1K concurrent connections at <200ms p99. 15+ sensor types (UART/I2C/SPI), industrial PLCs via Modbus RTU. Kalman filtering for sensor fusion, 10ms control loops at 99.9% uptime.
Go MQTT InfluxDB Modbus Embedded
2021 - 2022

Built data pipelines at Purplle processing 500M events/month from 7M+ MAUs.

ML Feature Store — BigQuery & Bigtable feature store reducing inference latency 85% (50s to 6s) for real-time pricing, enabling 8% conversion rate improvement.
BigQuery Bigtable
Go Migration — Migrated .NET monolith to Go microservices, cutting cloud costs $15K/month (30% memory, 25% CPU reduction).
Go Microservices
Self-Service DevOps — Terraform + GitHub Actions platform reducing provisioning from 2 hours to 3 minutes for 20+ teams.
Terraform GitHub Actions
2019 - 2021

Architected the AdTech platform at BookMyShow serving 5M+ MAUs. Led migration from PHP monolith to Go microservices, improving throughput 3x.

AdTech Platform — 100K ads/sec at <5ms p99 latency. Aerospike cluster with cross-DC replication achieving 99.99% uptime. CTR +25%, revenue $3M annually.
Go Aerospike Kafka
Priority Queue System — Kafka + NodeJS for customer support, reducing response time 80% (50min to 10min) for urgent requests.
Kafka NodeJS

other projects

AI & Research

Deep Agent System — LangGraph agents with planning, long-term memory, and tool-use via Claude API. Evaluated across 500+ test scenarios.
LangGraph Claude API Supabase
Talos Robotics — Vision transformers for robotic arm control with real-time object detection. Built ROS/Gazebo simulation and React telemetry interface.
PyTorch ROS Gazebo
LLaMA From Scratch — Full LLaMA 2 architecture in PyTorch: RMSNorm, rotary embeddings, grouped-query attention. 3x memory efficiency with gradient checkpointing.
PyTorch Transformers
Agricultural RAG — LLaMA + Ollama retrieval pipeline on 100K agricultural documents, serving 5K+ queries/month at 92% accuracy.
LLaMA Ollama LangChain

Web3 & Finance

Algorithmic Trading Platform — HFT strategy builder with backtesting engine processing 10M+ historical trades at sub-second simulation times.
Python Web3.py
DeFi Smart Contracts — Solidity contracts for automated trade execution and profit-sharing on Ethereum and BSC with $500K+ TVL.
Solidity Ethereum BSC
Real-Time Market Data — Aggregates prices from 20+ CEX/DEX sources, handling 50K events/second at <10ms latency.
Python Streaming