Hari Chenna profile photo 1 on Tal

Hari Chenna

Full-Stack & AI/Backend Engineer | Node.js, TypeScript, React | LLM/RAG Systems, vLLM, Kafka

Hyderabad

3+ years of work experience

Mar 20265 mosJul 2026

Remote

Refactored STT/LLM services using a factory architecture, removing unused API, WebSocket, and database components, reducing codebase size by 10%+ and improving maintainability and testability. Rearchitected STT/LLM services for the Makimoto AI open-source initiative by separating application logic from model-serving infrastructure, enabling independent deployments and reducing deployment time by 30%. Migrated LLM serving from Hugging Face Llama to vLLM, replacing sequential inference with continuous batching to double throughput and reduce latency by 30% under concurrent load, with an additional 12% gain from prefix-cache optimization. Built an LLM-as-judge evaluation framework to score response accuracy and relevance against reference answers, guiding iterative prompt and model improvements and cutting manual QA review effort by ~40%. Migrated STT inference from whisper-large-v3 to Faster-Whisper with CTranslate2, reducing inference latency by ~50%. Implemented Prometheus/Grafana observability across STT and LLM services, tracking inference latency, throughput, resource utilization, and service health to identify performance bottlenecks.
EdgeVerve
Jan 20251 yr 2 mosFeb 2026

Hyderabad, Telangana, India · Hybrid

Designed and developed a Commitment Fee Engine with a Node.js/Express backend and React-based configuration interface, supporting 4 fee types and using Kafka-driven events to synchronize fee data across a separate database while maintaining transactional consistency. Built full-stack Loan Split & Merge functionality with a Node.js backend and React frontend, using optimistic locking to eliminate race conditions during concurrent operations and ensure ACID-compliant data consistency. Built ATT (Automated Testing Tool), an automated regression-testing tool that cut regression testing effort by 70%. Resolved 150+ integration defects across distributed systems Mentored 4 junior developers through task planning, design reviews, and code reviews, reducing average onboarding-to-productivity time from 8 weeks to 5 weeks.
EdgeVerve
Dec 20222 yrs 2 mosJan 2025

Hyderabad, Telangana, India · Hybrid

Implemented role-based session authentication and authorization across 4 core banking APIs, strengthening access control for enterprise lending workflows. Developed a RAG Knowledge Retrieval Assistant (Python, LangChain, FastAPI, pgvector) to query banking API documentation and policy manuals, cutting average documentation lookup time by ~70% for internal teams. Built a PDF processing and embedding pipeline (chunking with overlap, pgvector storage) for semantic retrieval via cosine similarity, grounding LLM responses with citation guardrails and achieving ~85% Precision@5 on internal documentation retrieval.
GitHub
@Chari143
LeetCode
@Chari143

Hari’s AI Stack

claude
claude
github copilot
github copilot
langchain
langchain
ollama
ollama
openai
openai
vLLM
vLLM

Education

Bachelor of Technology - BTech, Electrical and Electronics Engineering

SAI RAJESWARI INSTITUTE OF TECHNOLOGY AND SCIENCES, LINGAPURAM

20193 yrs2022

Diploma of Education, Electrical and Electronics Engineering

Government Polytechnic College, Proddatur

20163 yrs2019

Join Hari and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Hari Chenna on Tal

Headline

Full-Stack & AI/Backend Engineer | Node.js, TypeScript, React | LLM/RAG Systems, vLLM, Kafka

About

Software engineer with 3.5+ years building production fintech and AI-powered systems — from React/Next.js interfaces down to Node.js/TypeScript backends, event-driven architectures, and LLM/RAG pipelines. I've doubled LLM-serving throughput through vLLM optimization, cut regression testing effort by 70% by building an automated testing tool, and reduced documentation lookup time by ~70% with a production RAG assistant — all within enterprise banking and fintech environments requiring ACID-compliant, concurrency-safe systems. Comfortable owning work end-to-end: API and schema design, event-driven microservices (Kafka), frontend interfaces (React/Next.js), and production AI infrastructure (RAG, embeddings, LLM-as-judge evaluation, inference optimization). Open to full-stack, backend, and AI/LLM engineering roles.

Location

Hyderabad

AI stack

claude

github copilot

langchain

ollama

openai

vLLM

Work experience

Senior Software Engineer (Contract – Client: Toku) at IT SPEAR, Mar 2026 - Jul 2026 · 5 mos, Remote, Refactored STT/LLM services using a factory architecture, removing unused API, WebSocket, and database components, reducing codebase size by 10%+ and improving maintainability and testability. Rearchitected STT/LLM services for the Makimoto AI open-source initiative by separating application logic from model-serving infrastructure, enabling independent deployments and reducing deployment time by 30%. Migrated LLM serving from Hugging Face Llama to vLLM, replacing sequential inference with continuous batching to double throughput and reduce latency by 30% under concurrent load, with an additional 12% gain from prefix-cache optimization. Built an LLM-as-judge evaluation framework to score response accuracy and relevance against reference answers, guiding iterative prompt and model improvements and cutting manual QA review effort by ~40%. Migrated STT inference from whisper-large-v3 to Faster-Whisper with CTranslate2, reducing inference latency by ~50%. Implemented Prometheus/Grafana observability across STT and LLM services, tracking inference latency, throughput, resource utilization, and service health to identify performance bottlenecks.

Product Developer at EdgeVerve, Jan 2025 - Feb 2026 · 1 yr 2 mos, Hyderabad, Telangana, India · Hybrid, Designed and developed a Commitment Fee Engine with a Node.js/Express backend and React-based configuration interface, supporting 4 fee types and using Kafka-driven events to synchronize fee data across a separate database while maintaining transactional consistency. Built full-stack Loan Split & Merge functionality with a Node.js backend and React frontend, using optimistic locking to eliminate race conditions during concurrent operations and ensure ACID-compliant data consistency. Built ATT (Automated Testing Tool), an automated regression-testing tool that cut regression testing effort by 70%. Resolved 150+ integration defects across distributed systems Mentored 4 junior developers through task planning, design reviews, and code reviews, reducing average onboarding-to-productivity time from 8 weeks to 5 weeks.

Associate product developer at EdgeVerve, Dec 2022 - Jan 2025 · 2 yrs 2 mos, Hyderabad, Telangana, India · Hybrid, Implemented role-based session authentication and authorization across 4 core banking APIs, strengthening access control for enterprise lending workflows. Developed a RAG Knowledge Retrieval Assistant (Python, LangChain, FastAPI, pgvector) to query banking API documentation and policy manuals, cutting average documentation lookup time by ~70% for internal teams. Built a PDF processing and embedding pipeline (chunking with overlap, pgvector storage) for semantic retrieval via cosine similarity, grounding LLM responses with citation guardrails and achieving ~85% Precision@5 on internal documentation retrieval.

Education

Bachelor of Technology - BTech, Electrical and Electronics Engineering, SAI RAJESWARI INSTITUTE OF TECHNOLOGY AND SCIENCES, LINGAPURAM

Diploma of Education, Electrical and Electronics Engineering, Government Polytechnic College, Proddatur

Public links

github: https://github.com/Chari143

leetcode: https://leetcode.com/Chari143