ACHYUTH
MACHAVARAPU · RÉSUMÉ
01 SUMMARY
02 EXPERIENCE
03 SKILLS
04 PROJECTS
05 ASK AI
06 EDUCATION
07 CONTACT
FULL-STACK DEVELOPER · GENAI & LLM APPLICATIONS

ACHYUTH MACHAVARAPU

I build AI-powered products that turn complex ideas into simple experiences.

CURRENTLY SHIPPING OHMSCHOOL.ORG ↗
READ THE RÉSUMÉ↓
2+
YEARS SHIPPING
PRODUCTION GENAI
3
ROLES ACROSS AI,
DATA & GEOSPATIAL
29
TECHNOLOGIES IN
ACTIVE USE
4
LANGUAGES SPOKEN
EN · TE · HI · TA
SCROLL
01 — PROFESSIONAL SUMMARY

Ambiguous requirements in.
Measurable AI features out.

Full-stack developer with 2+ years of experience shipping production-grade GenAI web applications end-to-end — React and Node.js/FastAPI platforms built around GPT-4o, Claude, and open-source LLMs. Hands-on with prompt engineering, RAG pipelines, fine-tuning, and automated evaluation; turns ambiguous requirements into reliable, measurable AI features.

02 — PROFESSIONAL EXPERIENCE

Three roles, one direction.

APR 2024 → PRESENT
  • Own GenAI features end-to-end across multiple products — prompt design, model selection, evaluation, and production rollout.
  • Built LLM-powered chat and agent workflows on GPT-4o, Claude, and open-source models; reduced hallucinations with structured prompting and guardrails.
  • Implemented RAG pipelines with Pinecone and pgvector to ground responses on proprietary knowledge bases.
  • Developed full-stack AI apps in React, Node.js, FastAPI, and Supabase with streaming responses and tool calling for real-time UX.
  • Fine-tuned models on domain datasets; automated evaluation with LLM-as-a-judge and regression suites to catch quality drift before release.
  • Validated and enriched large-scale geospatial datasets used to train mapping and routing AI models, raising spatial accuracy.
  • Ran multi-stage QA on digital map layers, resolving discrepancies to meet strict client SLAs; streamlined workflows to cut manual review effort.
  • Annotated NLP datasets for sentiment, NER, and intent classification, directly improving downstream model performance.
  • Authored tagging guidelines and led calibration sessions to raise inter-annotator agreement; built automation scripts on Amazon SageMaker Ground Truth to accelerate labeling throughput.
WHAT CHANGED BECAUSE I WAS THERE
Unbounded model output
Structured prompting, guardrails and LLM-as-a-judge regression suites — quality drift caught before release.
Ungrounded answers
RAG over Pinecone and pgvector — responses tied to the customer's own knowledge base.
Inconsistent labels
Written tagging guidelines and calibration sessions — higher inter-annotator agreement, better downstream models.
03 — TECHNICAL SKILLS

The stack, by weight.

LANGUAGES
04
Python · TypeScript · JavaScript · SQL
FULL-STACK
07
React · Node.js · FastAPI · Supabase · REST APIs · streaming UIs · tool/function calling
LLMS & GENAI
05
GPT-4o · Claude · Llama · OpenAI API · prompt engineering (CoT, few-shot, guardrails)
RAG & RETRIEVAL
05
LangChain · LlamaIndex · Pinecone · pgvector · embeddings & semantic search
FINE-TUNING & EVAL
04
LoRA · SFT · LLM-as-a-judge · regression test suites · hallucination reduction
CLOUD, DATA & NLP
04
AWS SageMaker (Ground Truth) · Git · data annotation · NER · sentiment · intent classification
04 — KEY PROJECTS & LIVE PRODUCT

Shipped, not prototyped.

FLAGSHIP

OHM School

DESIGNED, BUILT & SHIPPED END TO END
REACT · NODE.JS · FASTAPI · SUPABASE · LLMS

A modern K–12 learning platform built to make education personalized, interactive and measurable — curriculum, lessons, assessments, AI-powered learning, mastery, gamification, analytics and secure role-based experiences in one product.

12
MODULES
11
COMPONENTS
5
LAYERS
4
ROLES
P01

AI-Integrated Conversational Web Platform

REACT · NODE.JS · OPENAI API

Built a responsive web app with an embedded conversational AI chatbot for natural-language Q&A and task automation; designed prompt templates, conversation memory, and fallback handling for consistent on-brand responses.

P02

Video Content Annotation Pipeline for Vision AI

AMAZON SAGEMAKER · COMPUTER VISION

Designed the annotation strategy for object, action, and scene-context tagging; delivered high-precision labeled datasets that materially improved model reliability.

Inside OHM School — the architecture

11 COMPONENTS · 5 LAYERS · TAP ONE
PRESENTATION
ACCESS
DOMAIN
INTELLIGENCE
FOUNDATION
OHM SCHOOL STACK
Five layers, eleven components

Pick any component to see the part it plays. Every layer only talks to the one beneath it — security sits above the domain, not sprinkled through it.

05 — ASK MY RÉSUMÉ

Don't take my word for it. Interrogate it.

A live LLM, grounded only on this résumé — the same pattern I ship in production: structured prompting, scoped context, refusal when the answer isn't there.

GROUNDED ON THIS RÉSUMÉ · NO EXTERNAL CONTEXT
Ask about the stack, the roles, the RAG work, or whether he fits your opening.
06 — EDUCATION

B.E., Computer Science & Engineering

Hindusthan Institute of Technology — Coimbatore, Tamil Nadu
LANGUAGES
English · Telugu · Hindi · Tamil
AVAILABILITY
Open to remote and hybrid roles, worldwide
BASED
Los Angeles, CA (Remote) / India
07 — CONTACT

Let's build something
worth shipping.

EMAIL achyuthinsrirama@gmail.com PHONE +91 63034 40058 LIVE PRODUCT ohmschool.org ↗ OHM SCHOOL — BUILT & SHIPPED BY ME GITHUB FlicLabs/OHM-School ↗ THE PLATFORM SOURCE LINKEDIN machavarapu-achyuth ↗ PROFESSIONAL PROFILE
BASED Los Angeles, CA (Remote) / India UTC−7 / UTC+5:30
ACHYUTH MACHAVARAPU · RÉSUMÉ 2026 FULL-STACK · GENAI & LLM APPLICATIONS
PRESS / FOR THE TERMINAL
achyuth@resume — zsh
❯