Rajiv Yadav · R&D AI Engineer
I help companies deploy AI-powered backend systems that scale in production without compromising performance, reliability, or stability.
Trained, deployed and maintained in live systems - not notebook experiments.
Three multi-agent systems, nineteen agents. Every pipeline separates generation from review, so nothing ships on a single model call.
Three production RAG systems, each matched to the document type and retrieval pattern it serves.
I help companies deploy AI-powered backend systems that scale in production without compromising performance, reliability, or stability.
With deep expertise in machine learning infrastructure, distributed systems, and production-grade AI pipelines, I bridge the gap between research prototypes and reliable production systems. My focus is on building AI backends that handle millions of requests while maintaining sub-300ms latency.
Currently R&D AI Engineer, Bangalore, India.
What I do
End-to-end ownership - from architecture to deployment, monitoring, and the latency budget.
End-to-end design and deployment of ML models in production environments with high availability, low latency, and automated scaling.
Building robust ML pipelines, feature stores, model registries, and monitoring systems that keep your AI reliable at scale.
Architecting scalable LLM-powered applications with prompt management, caching, cost optimization, and multi-model routing.
Technical strategy and architecture reviews for companies building AI-first products. From prototype to production-ready systems.
Toolkit
The stack I reach for to take models from notebook to production - everything, upfront.
The journey
Roles where I shipped AI systems that had to stay up.
April 2026 - June 2026 · Bangalore, India
October 2025 - June 2026 · Gothenburg, Sweden
January 2026 - March 2026 · Mumbai, India
January 2025 - January 2026 · Noida, India
October 2023 - December 2024 · Chandigarh, India
Selected work
Real systems with real numbers - click any card for the full breakdown.
Developed a centralized gateway that routes requests across OpenAI, Anthropic, and Gemini through a single API interface. Built a monitoring dashboard for tracking latency, token consumption, request volume, error rates, and provider availability. The platform processed 12K+ requests, integrated 3 leading LLM providers, and exposed 10+ operational metrics to simplify experimentation and operational visibility for AI applications.
Built an agentic RAG platform with a 5-stage ingestion pipeline and 4-agent validation workflow, processing 5K+ document pages/day via OCR and supporting semantic retrieval over millions of vectors. Achieved ~45% reduction in hallucinated responses using validator + retry logic, with end-to-end query latency of ~5-8s. Optimizations reduced redundant embeddings by ~50% while improving retrieval accuracy and multi-mode query routing reliability
Built a production-grade Go authentication platform acting as a centralized identity provider for web and mobile apps, implementing JWT access (15m) and refresh (7d) tokens, device-bound sessions, RBAC, bcrypt password security, and Redis rate limiting. Delivered signup, OAuth, OTP, and recovery flows using PostgreSQL, Redis, AWS SES, and Twilio, enabling secure onboarding and session control for 50K+ users, supporting 1M+ monthly authentications, sub-25ms token validation, horizontal scalability, and instant global session revocation to reduce account takeover risks.
Built an institutional-grade trading engine combining 20+ TA features with a 5-model soft-voting ensemble (XGBoost, GBM, RF, SVM, MLP) for regime-aware signal generation. Implemented OHLCV ingestion, multi-timeframe resampling, SuperTrend filtering, ATR-based risk controls, and realistic backtesting with costs/slippage, achieving up to 39.4% returns and 62.5% win rates, outperforming buy-and-hold in volatile markets.
Built a real-time poll platform supporting 10K+ concurrent users with sub-200ms vote sync via WebSockets and stable handling of 100K+ votes/day. Service-repository architecture reduced feature delivery time by ~40%. API and DB optimizations cut redundant calls by ~60% and improved throughput 3x. System supports containerized deployment and horizontal scaling with seamless migration from SQLite to production databases.
Built a CNN-based pneumonia detection system trained on ~6,000 chest X-ray images using a 9-layer architecture, achieving 99.2% validation and 97% test accuracy, correctly classifying 1,200+ cases. Delivered as a scalable FastAPI backend with React frontend enabling real-time clinical inference and deployment-ready medical imaging workflows.
Walkthroughs of production systems I've shipped. Client work under NDA, so access is by request - send a quick note and I'll take it from there.
Build a matrix of companies against whichever metrics you want - revenue, profit, CEO commentary - pulled from their latest IR filings, every value carrying a citation. Exports to Excel, and generates an analysis of how each company is performing relative to the others.
Pick a topic and get the news on it as a daily or weekly digest, every item attributed to its source. Delivered both to email and inside the platform.
Pulls the Q&A section and the full financial statements - balance sheet, cash flow, income statement - out of every quarterly and annual filing. Statements download as Excel.
Social proof
Engineering leaders on what it's like to ship with me.
Rajiv has been a cornerstone in the development of the Test Case Genie and SQL Genie solution, driving key innovations such as LangGraph integration and parallel processing capabilities. His clarity of thought, rapid learning ability, and collaborative approach consistently elevate team performance and innovation.
Naveen Kumar Chauhan, Director, AI & Data Team, Axtria India, Axtria India
Rajiv consistently delivered impactful backend and AI infrastructure improvements at Karma AI, driving significant gains in system performance, reliability, and cost efficiency. His strong ownership, analytical mindset, and ability to solve complex engineering challenges made him a valuable contributor to the team and our platform's success.
Neeraj Tiwari, CEO & Founder, Karma AI, Karma AI
Rajiv made a strong impact by quickly identifying and resolving a critical AI infrastructure challenge, delivering a scalable solution that significantly improved system performance and efficiency. His technical expertise, problem-solving ability, and ownership mindset enabled him to contribute meaningful results in a short period of time. Rajiv is a capable AI engineer who consistently focuses on practical solutions and measurable business outcomes.
Sahng Ibrahim, CEO, EQL Global, EQL Global
[object Object]
[object Object]
[object Object]
AI and backend systems, under real load.
Grounding, four layers of execution safety, and the failure routing I'd defend hardest - assuming the test case might be wrong, not just the query.
We built the obvious thing first - point an LLM at a requirements doc, get SQL out. It worked, and we deleted it.
Bangalore, India