I engineer systems that think, reason, and scale — orchestrating AI agents, designing cloud architectures, and turning raw data into living infrastructure. From satellite imagery to LLMs, from published research to production pipelines.
I'm a computer scientist who refuses to stay in one lane. Born in India, educated at The George Washington University (3.9 GPA), and now operating at the intersection of cloud engineering, artificial intelligence, and data systems — I don't just build software, I build thinking infrastructure.
From mapping 10,000+ trees via ISRO satellite imagery to deploying multi-agent LLM pipelines in production, from engineering real-time IoT tracking systems to classifying 112,000+ chest X-rays — I bring research-grade rigor to production-grade code.
Currently an SDE at Core Software Technologies and AI Fellow at Handshake AI Solutions, where I architect cloud-native backends and validate the frontier of LLM behavior.
Dream. Aspire. Achieve. Repeat.
The DAAR Principle — my operating system.
Languages, frameworks, cloud services, and AI systems I wield to turn complex problems into deployed solutions.
Five roles across two continents — engineering, researching, mentoring, and building at the edge of technology.
Not theories. Not demos. Production-grade systems — from serverless data lakes to satellite imagery pipelines to autonomous agent frameworks.
A first-of-its-kind Autonomous Business Intelligence SaaS platform that converts any CSV upload into a full executive intelligence chain in under 3 seconds. Powered by a 7-stage autonomous ML pipeline: scikit-learn Isolation Forest (multi-metric anomaly detection across revenue, orders & AOV), Facebook Prophet forecasting with 95% confidence intervals, 9-dimension root cause attribution, playbook-driven recommendations, and Anthropic Claude natural language synthesis — all behind a FastAPI async backend and premium Next.js 16 frontend with 3D Framer Motion animations. Live-deployed on AWS App Runner + ECR at ~$8/month idle. 44 pytest cases. GDPR-compliant hard-delete API. Architected by Vivek Kommareddy.
A first-of-its-kind autonomous multi-agent AI control plane where 13 AI agents (4 Service, 4 Control, 5 Intelligence) negotiate compute, budget & SLA in a closed 15-phase loop — powered by 12 versioned Claude Skills (schema-validated LLM reasoning primitives: resource_request, negotiation, mediation, Shapley coalition, anomaly detection, compliance, carbon optimization, load forecast). Runs game-theoretic priority-weighted allocation + Shapley-value coalitions + Nash-convergent bid learning, a multi-cloud arbitrage engine (AWS/Azure/GCP cost+carbon+latency scoring), and TF-IDF vector memory with cross-tick strategy evolution. Live on AWS — S3 static site + FastAPI on Lambda (via Mangum) behind API Gateway + EventBridge-scheduled simulation processor persisting to DynamoDB. 40+ pytest tests, full React 18 + D3.js dashboard. Architected & built solo by Vivek Kommareddy.
A fully serverless AWS data engineering platform spanning 8 integrated services. Streams 200+ real-time order events through API Gateway → Lambda → Firehose into a 3-zone S3 data lake with under 2s end-to-end latency and zero dropped records. Automated schema discovery across 3 Glue catalog tables, JSON-to-Parquet transformation with 4 quality filters, cutting Athena scan cost by ~70%. Anomaly detection response time slashed by ~96% via EventBridge-scheduled Lambda.
End-to-end Retrieval-Augmented Generation system with vector embeddings, MMR retrieval, and cross-encoder reranking. Integrates OpenAI GPT, Anthropic Claude, and Ollama across 4 document formats. Hallucinations grounded via cited source chunks with confidence scores. Production-ready: REST API + Streamlit UI, 85%+ test coverage across 37 pytest tests, Docker containerized, CI via GitHub Actions.
Production-ready Python framework for autonomous deep research using a DAG of specialised AI agents — Planner, Researcher, Analyst, Writer, and Critic. Agents communicate exclusively via SharedState, never calling each other directly. Includes source credibility scoring, self-critique revision loops, vector semantic memory, exponential back-off retry, YAML declarative workflow composition, REST API with WebSocket progress feeds, and a Streamlit UI.
Full-stack iOS cricket auction app with 9 AI franchise agents, probabilistic decision-making, and real-time multiplayer. Processes 2,160 bid decisions in under 14 seconds across 240 verified IPL 2026 players.
AI-powered tool that analyzes resumes against job descriptions to compute match scores, identify missing skills, and generate targeted improvement suggestions.
Detected and mapped 10,000+ trees using 120+ Cartosat-2 satellite images (ISRO). CLAHE + multi-Otsu thresholding + OpenCV — 92.7% accuracy. Boosted processing speed 45% via QGIS tiling.
Processed 3,000+ retinal fundus images to detect diabetic retinopathy indicators — 94.1% accuracy. CNN-based DR classification with Python GUI, enabling screening for 250+ patients across 2 hospital trials. Resulted in a granted patent.
Analyzed 112,120 chest X-rays from 30,805 NIH patients. InceptionResNetV2 vs VGG19 comparative study — achieved 94.2% accuracy classifying 5+ lung conditions. Outperformed VGG19 by 6.4%.
A curated collection of graduate-level ML and statistical modeling implementations — probability distributions, SVM, CNNs, Markov models — using Jupyter Notebooks with in-depth analysis.
A patent and two Scopus-indexed journal papers. Research that moved from laboratory to published record.
A CNN-based system detecting optic disc, exudates, and microaneurysms from retinal fundus images — enabling automated screening for diabetic retinopathy with 94.1% accuracy.
Whether you're solving a hard problem, building a team, or just want to talk about AI agents and cloud architecture — I'm listening.