Building backend services and applied AI systems that hold up under real load: enterprise REST APIs on SQL/PostgreSQL backends, deterministic state machines for orchestration and recovery, simulation-guided decision agents, and embedding-based retrieval pipelines engineered for correctness and latency at scale.
I'm a Computer Science undergraduate at VIT Vellore, and most of what's below is a record of moving systems past the prototype stage replacing fragile, manual workflows with services that degrade gracefully and recover on their own.
BKT — Software Engineering Summer Intern · 10 May 2026 – 30 June 2026
Worked alongside SAP teams and business consultants to replace manual, repetitive aggregation queries with a robust API layer for internal sales analytics.
- Developed a Spring Boot backend exposing REST APIs over SQL Server data for recurring sales analytics.
- Wrote SQL Server stored procedures, functions, and triggers to support enterprise sales workflows.
- Validated OutSystems CRUD operations to protect data integrity across the platform.
- Built Power BI dashboards and SQL reports to monitor sales KPIs for business reporting.
ASTra — A codebase comprehension tool that parses Java abstract syntax trees into a searchable, semantic index queried in plain English.
Java 21 Spring Boot PostgreSQL pgvector JavaParser LLMs
View details
- Architected a Spring Boot REST service backed by PostgreSQL with the pgvector extension for approximate nearest-neighbor search over embedded code symbols.
- Built an adaptively batched, multithreaded indexing pipeline that reduced processing time by ~49% when parsing a 1,700+ class repository.
- Implemented semantic chunking and token-budget-aware context assembly to feed multiple LLM backends (OpenAI, Ollama, Groq), returning answers with exact line-level source references.
- Built a decoupled, interactive CLI client and an automated benchmarking suite to measure retrieval-quality and latency regressions.
Multi-Threaded DPI Engine — A high-performance network traffic analysis platform combining a multi-threaded C++ deep packet inspection engine with a Spring Boot orchestration backend and AI-powered insights.
C++ Spring Boot Redis Docker Llama 3
View details
- Architected a multi-threaded C++ DPI engine with a load-balancer → fast-path worker pipeline, using five-tuple flow hashing to route packets deterministically and keep per-connection state race-free across threads.
- Built a layered packet-parsing pipeline (Ethernet/IP/TCP/UDP) with TLS SNI and HTTP host extraction to classify traffic by application and domain, applying rule-based blocking and emitting structured JSON analytics.
- Designed a Spring Boot orchestration layer that runs the native binary asynchronously via
ProcessBuilder, with Redis-backed job-state and AI-response caching, and Groq (Llama 3.3-70B) generated traffic risk summaries.
ResilientDB — A pluggable backup orchestration engine that decouples core scheduling logic from database-specific backup implementations.
Java Spring Boot PostgreSQL AWS (EC2, RDS, ECR) Docker GitHub Actions
View details
- Uses a Strategy Pattern-based engine supporting interchangeable PostgreSQL, MySQL, and MongoDB backup drivers without touching core orchestration code.
- Runs an asynchronous worker pool with exponential backoff retry logic to prevent HTTP thread-pool exhaustion during sustained backup load.
- Models job lifecycle as a deterministic finite state machine (PENDING → PROCESSING → COMPLETED/FAILED) with real-time health metrics, cutting mean time to recovery from hours to seconds.
| Project | What it does | Stack |
|---|---|---|
| Eco-Looping Building Agent | Closed-loop control system that optimizes HVAC energy consumption and occupant comfort through simulation-guided decision-making, with a Groq Llama 3.3 agent evaluating a matrix of simulated physical outcomes rather than fixed temperature heuristics. | Python EnergyPlus Groq Llama 3.3 Streamlit Pydantic |
| TranscriptIQ | Fault-tolerant audio intelligence pipeline turning long-form podcast transcripts into precisely retrievable answers, with a multi-tier ingestion pipeline cutting speech-to-text latency by 85% and FAISS + Llama-3.3 retrieval cutting search time by 65%. | Python Transformers FAISS Whisper Llama-3 via Groq |
| Email Writer Service | Chrome extension that injects AI-generated, tone-controlled reply drafts directly into Gmail's compose UI via a stateless Spring Boot + Gemini API backend. | Java Spring Boot React Chrome Extension Gemini API |
| Hybrid Lexical-Semantic Matching | Interpretable NLP pipeline for candidate-role matching, combining TF-IDF lexical scoring with MiniLM embedding similarity and an explainability layer that surfaces top-contributing sentences behind each ranking. | Python Scikit-learn Sentence Transformers Streamlit |
View certifications (5)
Microsoft Certified: Azure AI Engineer Associate — Microsoft · Issued Jul 2025 · Expires Jul 2027
Oracle Fusion AI Agent Studio Certified Foundations Associate — Oracle · Issued Oct 2025
Oracle Data Platform — Oracle · Issued Oct 2025
Oracle AI Cloud Infrastructure — Oracle · Issued Oct 2025
Oracle AI Foundation — Oracle · Issued Oct 2025