Rajiv Ranjan Kumar
>
📍 Jaipur, IndiaOpen to challenging AI/ML work · Remote or on-site across India
Areas of focus
RAG & Retrieval
Hybrid search, reranking, vector DBs, text-to-SQL
Model Training & Fine-tuning
LoRA/PEFT, VLMs, quantization, classical ML
Architecture & Systems
Multi-agent orchestration, LLM serving at scale
AI Projects
End-to-end products shipped to real users
LLM Security & Red-teaming
OWASP LLM Top 10, jailbreaks, guardrails, evals
Selected work
Tool-Selection LLM — LoRA Fine-tune on xLAM
Teaching a small LLM to pick the right tool/API for an agent — and say how confident it is. LoRA-tuned Qwen3.5-2B with Unsloth on xLAM function-calling data: 50.6% → 100% accuracy, ECE 0.220 → 0.002.
DocuMind AI — Grounded Multimodal Document Q&A at Scale
System design for a 1M-user document Q&A product: async multimodal ingestion, hybrid retrieval, a deterministic confidence gate that refuses instead of hallucinating, and hot/warm/cold session tiering.
Agentic AI Chatbot Platform
Production multi-agent assistant for cybersecurity & case management — a router agent dispatching to five specialist agents across hybrid RAG, text-to-SQL and action execution, on self-hosted vLLM.
RedForge — AI Red-teaming & GenAI Eval Platform
Target-agnostic platform that attacks any LLM chatbot with 200+ OWASP-LLM-Top-10 test cases and an adversarial mutation engine. Built on personal time, adopted as a company product.
3B Tokens/Day Vision-LLM Inference on One GPU
Scaled self-hosted vision-LLM screenshot analysis to 3 billion input and 50 million output tokens per day on a single 96GB GPU — no third-party API spend.
Skill clusters
Agentic & GenAI
LLM Serving & Tuning
Retrieval & Vector DBs
Evaluation & LLM Security
ML & Data Science
Backend & Deployment
Cloud & Observability
Automation & BI
Experience
- Aug 2026 — Present
AI / Data Science Developer, SDE-II — R&D
Izoologic — Cybersecurity & Digital Threat Management · Jaipur, IndiaLeading agentic AI platform work and the RedForge red-teaming product.
- Lead RedForge — AI red-teaming platform adopted as a company product
- Architect of the multi-agent assistant serving 400+ enterprise clients
- Aug 2025 — Aug 2026
AI / Data Science Developer, SDE-I — R&D
Izoologic — Cybersecurity & Digital Threat Management · Jaipur, IndiaBuilt production LLM systems end to end.
- Multi-agent chatbot on LangGraph + self-hosted vLLM: 200 concurrent users, 500–600 concurrent LLM requests
- Self-learning few-shot engine: text-to-SQL accuracy 60% → 95%
- Vision-LLM inference scaled to 3B input tokens/day on a single 96GB GPU
- LoRA/PEFT fine-tuned VLM for automated threat triage
- May 2024 — Jul 2025
Business Intelligence Developer
Genpact · Jaipur, IndiaForecasting and automation for a leading Australian banking client.
- Predictive forecasting models beating the existing baseline
- Automated reporting & reconciliation — 40% less manual effort
- Power BI / Tableau KPI dashboards adopted across finance & ops
- Jan 2023 — Apr 2024
Data Science Intern
AI Variant · Bangalore, IndiaApplied ML and NLP.
- Recommendation systems and NLP sentiment pipelines (scikit-learn, pandas, NLTK)
- EDA and data storytelling to guide product decisions
Learning log
Accuracy isn't enough: calibrating a tool-selection model
Fine-tuning Qwen3.5-2B to pick tools took accuracy from 50.6% to 100% — but the number I cared about was ECE, 0.220 → 0.002.
Hybrid search in practice: BM25 + dense + reranking
Why dense-only retrieval fails on identifiers and jargon, and how fusing sparse and dense results before a cross-encoder fixes it.
OWASP LLM Top 10 — field notes from building a red-team harness
What actually breaks production chatbots: indirect injection, system-prompt extraction and excessive agency.
Let's build something intelligent.
AI Engineer (SDE-II) with 3+ years across applied machine learning, data science and production LLM systems. I build agentic AI platforms end to end: multi-agent orchestration, hybrid RAG, text-to-SQL, tool/function calling and self-hosted inference — including vision-LLM pipelines scaled to 3B input tokens/day on a single 96GB GPU. I care about the full stack: context engineering, retrieval and evaluation through to containerised deployment, observability and cost/latency optimisation. I also built RedForge, an AI red-teaming platform that went from a self-initiated side project to an adopted company product. Domain experience in cybersecurity, threat management and BFSI.