
AI Engineer with less than a year in RAG pipelines and multi-agent orchestration.
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
M.Tech AI candidate at COEP (2025–27) building production-grade LLM systems — RAG pipelines with 93% answer relevance, LangGraph multi-agent orchestration, and FastAPI NLP backends with P95 tracking. Focused on evaluation-driven development and building AI systems that are measurable, reliable, and production-minded.
COEP Technological University
M.Tech · Artificial Intelligence
August 1, 2025 – Present
Vidyalankar Institute of Technology
B.E. · Information Technology
August 1, 2020 – June 30, 2024
Intelligent Customer Support Ticket Routing System
June 1, 2026 – Present
Fine-tuned RoBERTa-base on 12K customer support tickets across 8 intent classes, achieving 75% accuracy and 0.76 macro F1; replacing rule-based keyword routing with async multi-intent inference. Built an asynchronous FastAPI backend with Redis caching, confidence-threshold routing, IP-based rate limiting (10 requests/min), and latency observability with P95 performance tracking. Integrated Groq-hosted LLaMA 3.1-8B for intent-aware response generation, implementing safety guardrails for PII protection, payment-retry prevention, and sentiment-based ticket prioritization. Developed a real-time Streamlit analytics dashboard featuring intent distribution, confidence-score monitoring, and latency analytics for operational insights.
View ProjectSentinel: RBAC Secured RAG System
June 1, 2026 – Present
Architected an RBAC-enforced RAG platform with access control applied directly at the Qdrant vector layer, ensuring unauthorized document chunks never reached the LLM. Engineered a multistage retrieval pipeline leveraging HyDE, hybrid BM25+dense retrieval, RRF fusion, and cross-encoder reranking, achieving 93.8% Answer Relevance. Achieved 87.7% Faithfulness through LLM-as-a-Judge evaluation, leveraging confidence-calibrated decision gating and sliding-window context expansion to reduce unsupported responses. Designed a confidence-calibrated decision gate (hard-hit, soft-hit, no-info) achieving 100% gate accuracy while eliminating unnecessary LLM calls on low-confidence retrievals.
View ProjectDelphi: Self-Improving Multi Agent Decision Intelligence Platform
June 1, 2026 – Present
Architected a LangGraph-based multi-agent reasoning system that dynamically assembles 4-8 specialized expert councils per query using dual-model inference and debate-driven decision making. Engineered self-healing agents with ELO-based reputation tracking and recovery mode, injecting historical failure/success lessons and enforcing domain-specific self-critique for underperforming experts. Evaluated the framework across 75 simulated domain scenarios, improving confidence calibration by +0.25 while maintaining stable expert reputation dynamics (4.66 ELO drift). Built an asynchronous FastAPI backend with aiosqlite persistence and validated system reliability through 45 unit and integration tests using isolated in-memory databases.
View ProjectCultural Fit Analysis
The candidate's projects demonstrate a proactive and innovative approach to AI challenges, aligning well with a culture that values research, development, and practical application of cutting-edge technologies. The diversity of projects (RAG, multi-agent systems, NLP for customer support) indicates adaptability and a broad interest in AI applications. The pursuit of a Master's degree in AI while actively building complex systems shows a commitment to continuous learning and growth.
Soft Skills & Operational Fit
The candidate's project descriptions indicate a strong problem-solving aptitude and a methodical approach to system design and evaluation. The emphasis on 'measurable, reliable, and production-minded' AI systems suggests a good operational fit for roles requiring robust and maintainable solutions. The detailed project descriptions also imply strong written communication skills.