AI Engineer
ML Engineer
Data Scientist
Full-Stack
QA Automation
Building production-grade AI systems from LLM routers to enterprise RAG pipelines.
10 shipped projects · 97% cost reduction · GPA 8.1
I'm a final-year B.Tech Computer Science & Data Science student at Vidya Jyothi Institute of Technology, Hyderabad — graduating May 2026 with a GPA of 8.1.
I've shipped 10 production-grade projects spanning enterprise RAG pipelines, LLM fine-tuning systems, full-stack AI analytics platforms, machine learning pipelines, and QA automation frameworks — work that would conventionally take months, delivered with precision and measurable impact.
My philosophy: AI is a force multiplier, not a shortcut. I've engineered LLM routers that cut API costs by 81%, synthetic data pipelines that fine-tune models at 97% lower inference cost, and production RAG systems with PII guardrails, confidence scoring, and Docker deployment — all as a fresher.
Actively seeking entry-level roles in AI Engineering, Data Science, Software Engineering, and QA Automation.
ENTERPRISE-GRADE 7-LAYER PIPELINE
Enterprise RAG system with PII auto-redaction via Presidio (10+ entity types), input/output safety guardrails, retrieval confidence scoring with graceful fallback, and a full analytics dashboard. React 18 frontend + async FastAPI + ChromaDB/pgvector backend — fully Dockerized.
View on GitHubPRODUCTION BENCHMARK PLATFORM
Benchmarks LLM outputs across quality, cost, latency, and hallucination rate with a live Grafana dashboard and FastAPI backend. Compare models side-by-side with real-time metrics.
LoRA / QLoRA FINE-TUNING PLATFORM
Fine-tunes open-source LLMs (Mistral-7B, LLaMA-3) on domain-specific data using LoRA/QLoRA. Full dataset manager, training dashboard, benchmark comparison, and inference playground.
97% LOWER INFERENCE COST
Uses a large LLM as teacher to generate training data, then fine-tunes a tiny model via LoRA/PEFT+TRL on Phi-3-mini with 4-bit quantization — the tiny model beats the large one at 97% lower cost.
SYSTEM 1 VS SYSTEM 2 ROUTING · 81% COST CUT
Scores prompts across 7 weighted complexity dimensions, auto-routing simple tasks to cheap/fast models and complex reasoning to premium ones. Supports Anthropic + OpenAI interchangeably. Zero accuracy tradeoff.
PRODUCTION RAG · REACT 18 + FASTAPI
Full production RAG pipeline: 512-token semantic chunking → all-MiniLM-L6-v2 embeddings → ChromaDB retrieval → Claude API with citations. Async FastAPI + SSE streaming + React 18 frontend with drag-and-drop, live token streaming, and relevance scores.
AI E-COMMERCE ANALYTICS PLATFORM
Full-stack data science platform over 8,000+ synthetic orders: Random Forest churn (AUC 0.658), Gradient Boosting revenue forecast, Apriori market basket, Claude API as natural language Business Analyst, and 10 optimised SQL queries powering real-time KPI cards.
END-TO-END ML PIPELINE · ROC-AUC 0.754
3-model pipeline on 5,000 telecom records with full feature engineering. Best ROC-AUC 0.754 with Gradient Boosting. SHAP-style importance analysis. Fully reproducible EDA-to-deployment pipeline in scikit-learn.
RELATIONAL DATA ENGINEERING · 17 QUERIES
Fully normalised 4-table relational schema with 17 analytical queries — JOINs, correlated subqueries, RANK/DENSE_RANK window functions, CASE WHEN logic. Python-SQLite execution layer auto-generates 5 Matplotlib charts.
MULTI-COUNTRY PANDEMIC ANALYTICS
Structured EDA across 20 countries and 36 months, engineering 5 publication-quality visualisations covering case trajectories, vaccination rollout impact, regional death-rate disparities, and socioeconomic correlations.
Open to full-time roles in Data Science, AI Engineering, Software Engineering & QA Automation. Based in Hyderabad — open to relocation anywhere.