Headline

AI Engineer — Production GenAI & Agentic Systems | RAG · LLM Orchestration · AI Reliability/Evaluation · Python, FastAPI

About

I'm an AI Engineer who builds production GenAI and agentic systems end-to-end — architecture, LLM/agent orchestration, RAG, evaluation, backend integration, and deployment. At International Media Company (IMC), I independently designed and shipped Caption-Generator, an AI content pipeline (transcription → hallucination detection → vision fallback → multilingual copy generation → compliance validation → human review) that's now in daily production use by editors, processing 50–100 clips a day on our own infrastructure with automated scheduling. I'm also building FunnelCore, a configuration-driven attribution platform for tracking acquisition sources and paid signups. Before IMC, I spent close to a years at OctoGenie building a multilingual Legal AI platform — architecting an orchestration layer across 8+ services (RAG, STT, TTS, translation, court scraping, lawyer verification), engineering a Legal RAG system across 9+ Indian languages, and hardening scraping infrastructure across 10+ state court systems. I started my career in ML at CyborgIntell, building and deploying credit-underwriting and decisioning models that cut default rates and improved approval efficiency in a live lending workflow. What I care about: making probabilistic systems trustworthy in production — deterministic controls, evaluation harnesses, observability, and audit trails, not just prompts. Currently focused on: Agentic AI · RAG · LLM Orchestration · AI Reliability & Evaluation · AI Automation 📫 jkridutta@gmail.com | github.com/jahnvik16

Location

India

AI stack

claude code

cursor

notebooklm

Exa

Jina AI

Stable Diffusion

Work experience

Artificial Intelligence Engineer at IMC, Jun 2026 - Present · 4 mos, Remote, Architected and independently shipped Caption-Generator, an end-to-end AI content pipeline (transcription → hallucination/timing detection → GPT-4o-mini vision fallback → multilingual copy generation → compliance validation → human review) for Instagram, TikTok, and X — deployed on the company's own server with an automated scheduler queuing and dispatching clips, in daily production use by editors at 50–100 clips/day. Engineered failure-safe pipeline architecture with per-stage error isolation, asynchronous background job processing, and feedback-driven regeneration that preserves prior successful output; verified correctness via 148 automated tests plus a manual browser QA pass across the full clip lifecycle. Designed and built FunnelCore, a configuration-driven attribution platform (Next.js, PostgreSQL, Prisma) modeling click-time attribution, immutable tracking-link versioning, and Telegram/aggregator routing with AES-256-GCM credential encryption and full audit logging — delivered V1 through Phase 7 and staged for production traffic onboarding.

AI Engineer at OctoGenie, Feb 2026 - Jul 2026 · 6 mos, Bengaluru, Building production AI systems across LLM orchestration, Legal RAG, and multilingual voice pipelines. Focused on reliability, observability, and failure-resilient architecture for real-world deployment. Key work: - Designed a hybrid LLM + deterministic tool routing layer across RAG, scraping, verification and voice workflows - Built a Legal RAG system with FAISS retrieval and multilingual support for 9+ Indian languages - Engineered real-time court data pipelines with CNR routing, retry/backoff, and proxy integration - Instrumented pipelines with Langfuse for end-to-end observability and structured failure classification

Artificial Intelligence Engineer at OctoGenie, Aug 2025 - Jun 2026 · 11 mos, Architected a production AI orchestration layer unifying 8+ services (RAG, STT, TTS, translation, court scraping, lawyer verification) through deterministic tool-routing and standardized JSON contracts — establishing integration patterns adopted across the backend team. Engineered multilingual Legal RAG (FAISS + IBM Granite embeddings, 9+ Indian languages) and a Whisper large-v3 → RAG → Meta MMS-TTS/Edge-TTS voice pipeline; resolved CUDA OOM/cuDNN failures via CPU fallback and isolated malformed translation-layer JSON from model regressions. Hardened scraping and Bar Council verification infrastructure across 10+ state court/portal systems (retry/backoff, idempotent cron guards, unified JSON schema across 8+ portals); resolved two production bugs eliminating silent data loss and benchmarked 4 TTS candidates to remove 2 underperformers.

AI Engineer Intern at OctoGenie, Aug 2025 - Jan 2026 · 6 mos, Bengaluru, Built foundational AI infrastructure including a multilingual voice pipeline and a multi-state lawyer verification system used to ground LLM responses. Key work: - Built STT → RAG → TTS voice pipeline supporting 9+ Indian languages using Whisper and MMS/Coqui TTS with fallback strategies - Optimised inference latency using async FastAPI processing and request batching - Developed a Bar Council verification API scraping and normalising multi-state lawyer data for hallucination-resistant LLM grounding

Data Analyst at CyborgIntell, Jan 2025 - Aug 2025 · 8 mos

Data analyst intern at CyborgIntell, Jun 2024 - Dec 2024 · 7 mos

Education

Bachelor of Engineering, Computer Engineering, Savitribai Phule Pune University

Public links

github: https://github.com/jahnvik16