AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
AI Engineer with less than a year in GenAI and RAG pipelines.
GenAI Engineer with hands-on production experience building RAG pipelines, multi-agent LLM systems, and scalable AI microservices—from architecture through deployment and evaluation. Shipped a live LangChain + FAISS RAG system (NutriLens), a production agentic recommendation engine (Intentra), and an MLOps-automated emotion detection platform (SMOOD) with Apache Airflow retraining. At Codec Technologies, reduced inference latency 30% via async optimisation and ONNX, sustaining 200+ concurrent requests in load tests. Proficient across the full modern AI engineering stack: multi-agent orchestration (LangGraph, AutoGen, Google ADK, MCP), LLM evaluation (RAGAS, LangSmith), retrieval systems (Pinecone, Qdrant, Graph RAG, rerankers), model fine-tuning (LoRA/QLORA), and inference serving (vLLM, Ollama) combined with production-grade FastAPI backends, Docker, and CI/CD. Deeply interested in multilingual NLP, distributed AI systems, and building reliable AI products at scale.
Dr. A.P.J. Abdul Kalam Technical University
B.Tech · Artificial Intelligence
August 1, 2022 – June 30, 2026
Codec Technologies
AI/ML Developer Intern
October 1, 2025 – November 1, 2025
India
SafeScape Community Safety Navigation Platform
June 1, 2026 – Present
Led as backend and AI developer; delivered an offline-first Progressive Web App (Service Workers + local cache) enabling safety map access in zero-connectivity environments—relevant to low-bandwidth, vernacular-user deployment contexts. Built a Python AI safety scoring layer aggregating community incident reports into street-level risk scores, feeding a Leaflet.js heatmap with hyperlocal resolution down to individual street corners.
View ProjectSMOOD Real-Time Emotion Detection Platform
June 1, 2026 – Present
Trained a ResNet-18 emotion classifier on FER-2013 (35k images, 7 classes); achieved 68% validation accuracy—a 15 percentage-point improvement over MobileNet-V2 baseline—through targeted augmentation (random horizontal flip, rotation, colour jitter) and learning-rate scheduling; documented performance ceiling as label noise, not model capacity. Built an Apache Airflow DAG scheduling nightly retraining (data ingestion → training → evaluation → model registry push), eliminating manual retraining overhead—production MLOps discipline applied to a real CV pipeline.
View ProjectNutriLens Production RAG Nutrition Intelligence System
June 1, 2026 – Present
Architected and shipped to production a LangChain RAG pipeline combining OCR-extracted food label data with a FAISS vector store (10k+ item corpus); publicly live at nutrilens-r5we.onrender.com. Engineered few-shot prompt templates with Pydantic output parsers enforcing structured JSON from GPT-4, eliminating unstructured hallucinations in nutritional fields; applied RAGAS-style evaluation (Faithfulness, Answer Relevance, Context Precision) across 500+ test queries to validate retrieval quality and response consistency. Implemented cosine similarity thresholding on retrieved chunks—returning a graceful "insufficient data" response rather than a hallucinated answer when no relevant context exceeds the confidence floor. Containerised with Docker and deployed via Render; FastAPI backend exposes /analyse and /history endpoints with JWT-secured user sessions and full OpenAPI documentation.
View ProjectIntentra Intent-Aware Agentic Recommendation Engine
June 1, 2026 – Present
Designed a hybrid agentic + deterministic architecture implementing a ReAct-inspired separation of concerns: Gemini API extracts structured intent from free-text queries (LLM as semantic layer); a deterministic scoring engine (rating × distance × open status) ranks results — eliminating hallucination risk from the execution layer entirely. Implemented production-grade middleware: X-Request-ID context propagation, X-Process-Time-Ms logging, and a 504 timeout guard — full observability stack for on-call ownership. Shipped a complete CI/CD workflow (ruff → pytest → Docker build); authored Dockerfile, .dockerignore, dependency-pinned requirements.txt, and centralised app/core/config.py with typed Pydantic config. Delivered advanced agent API features: surprise_mode, open_now_only, max_distance_km filters; /vibes catalog; /health liveness probe; and matchup head-to-head explanation field — all with tool-use patterns and structured function outputs.
View ProjectGoogle Cloud Generative AI Certification
Google Cloud
June 1, 2026 – Present
J.P. Morgan Software Engineering Virtual Experience
J.P. Morgan
June 1, 2026 – Present
Cultural Fit Analysis
The candidate's project diversity (community safety, emotion detection, nutrition intelligence, recommendation engine) and engagement in a hackathon demonstrate initiative and a broad interest in applying AI to various domains. Their focus on building 'reliable AI products at scale' and interest in 'multilingual NLP, distributed AI systems' aligns well with an innovative and globally-minded engineering culture. The certifications and ongoing education (German language) also indicate a commitment to continuous learning and self-improvement.
Soft Skills & Operational Fit
The candidate demonstrates strong communication skills through detailed project descriptions and experience in authoring runbooks and OpenAPI documentation. Their experience in Agile sprints and cross-functional team collaboration indicates good operational fit. The mention of 'Async-first Collaboration' and 'documented decision logs' suggests a proactive approach to team communication and knowledge sharing, which is crucial for distributed teams.