Skip to content
View Epik-Whale463's full-sized avatar
🎯
Focusing
🎯
Focusing
13 followers · 9 following
  • 01:54 (UTC +05:30)

Achievements

Achievement: QuickdrawAchievement: Pull SharkAchievement: YOLO

Achievements

Achievement: QuickdrawAchievement: Pull SharkAchievement: YOLO

Organizations

@AIHUB-VVIT

Block or report Epik-Whale463

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Epik-Whale463/README.md

Rama Charan

AI/ML Engineer · Bengaluru

Building applied AI for Indic languages — LLM pipelines, evals, and the backend that holds them up in production. Recently shipped Setu V2 at Pratilipi, a multi-stage LLM adaptation pipeline now in org-wide review for rollout across 12 Indic languages.

Reach me: rama.charan.official@gmail.com · LinkedIn · Resume

What I work on

  • Multi-stage LLM pipelines with evals — decomposition over monolithic prompting
  • Multilingual NLP for Indic languages (translation, classification, TTS)
  • Backend that makes ML actually ship — FastAPI, async, auth, deployment
  • Eval and guardrail design for non-deterministic systems

Selected work

Setu V2 (private, Pratilipi) — 3-pass source-blind LLM adaptation pipeline replacing a single-pass V1. Authored a 7-dimension evals benchmark; 58% fewer adaptation defects on blind A/B with LLM-as-judge validated against human reviewers. In org-wide review for rollout.

ContentClosureFlow (private, Pratilipi) — Multilingual author-abandonment classifier (multilingual-e5-large + logistic regression) productionized as a daily AWS Step Function. 90.5% cross-lingual accuracy after diagnosing and fixing a label-leakage issue that took the first-pass score from 44.5%.

KhetAI — Tool-augmented LLM assistant for Indian agriculture queries. Intent-classified routing across 6 live API tools (weather, mandi prices, plant disease ID, geolocation, news, LLM). Validated with 10 farmers near my college. Live as an Android APK.

SpeechT5 Telugu TTS — Fine-tuned SpeechT5 on 8,576 IndicTTS Telugu samples using a Telugu-to-Latin transliteration shim so the English-trained tokenizer could process Telugu. Validation loss 0.67 → 0.45. Open-sourced on Hugging Face; 600+ downloads. Loss-based evaluation only — see model card for limitations.

Stack

Python, TypeScript · PyTorch, Hugging Face, scikit-learn · FastAPI, Next.js, async · AWS (Step Functions, S3, ECR), GCP · PostgreSQL, MySQL, MongoDB, Qdrant · Auth0, SQLAlchemy

Background

B.Tech AI/ML, VVIT (2026) · AI/ML Engineer Intern at Pratilipi (Feb-Jun 2026, Firebolts team) · Full-stack AI Intern at AI4Bharat / IIT Madras (Oct 2025 – Feb 2026)

AWS Certified AI Practitioner · NPCI Bharat AI Quest 2025 finalist · ISRO Bharatiya Antariksha Hackathon 2025 shortlist

Pinned Loading

  1. ChatWithPdf ChatWithPdf Public

    Python

  2. College_Query_System College_Query_System Public

    ACCESS THE LIVE DEPLOYMENT HERE

    HTML

  3. GitManager GitManager Public

    My project for GitManager

    Shell

  4. retrieval-augmented-generation-for-Cancer-data retrieval-augmented-generation-for-Cancer-data Public

    A RAG model with cancer data awareness

    HTML