SOUJANYA GUPTA

Summary

I’m a software engineer who works on evaluating AI agents and large language models. My work grows a WildClawBench fork on OpenClaw into a multi-provider evaluation platform with three-judge council grading, drift injection across 101 mock services, bounded multi-agent orchestration, and cost tracking for Claude on AWS Bedrock, GPT-5.5, and OpenRouter. Alongside that, I build production frontend platforms for MyMandi and Kuberha, including operations dashboards, app integrations, Shiprocket delivery workflows, and AI features, as part of Ethara AI’s engineering team.

Professional experience

Ethara AI (formerly Green Rider Technology) · Gurugram, India

Agent evaluation platform

Expanded a WildClawBench fork on OpenClaw into a multi-provider platform for Claude on AWS Bedrock with extended thinking, GPT-5.5, and OpenRouter.

• Built a LiteLLM sidecar for prompt caching and per-request cost accounting.
• Reworked grading into a three-judge council with separate evidence budgets and partial-coverage majority voting, alongside weighted pytest suites.
• Built drift injection across 101 FastAPI mock services using a mutable store, hidden admin API, and host-side director.
• Added up to 5 concurrent sub-agents, each limited to 50 tool calls, 16k tokens, and 600 seconds, with spawn audit trails.
• Authored 13-turn persona tasks with silent mutations and confirmation gates; delivered Docker isolation, offline dependencies, Whisper transcription, and multimodal trajectories on EC2 with Git LFS.

Model & dataset quality

Built evaluation and data-quality workflows for model training across finance, e-commerce, and healthcare.

• Wrote domain prompts and evaluation rubrics for a Meta LLaMA 4 Maverick fine-tuning program delivered to a leading AI data platform.
• Built Python and LangChain pipelines to benchmark OpenAI, Gemini, Claude, LLaMA, and Grok.
• Owned taxonomy classification, entropy checks, inter-rater agreement, NLP preprocessing, and structured annotation.
• Built an internal Odoo application for labeling golden datasets.

Frontend platforms: MyMandi & Kuberha

Worked as a frontend engineer within Ethara AI’s engineering team on MyMandi and Kuberha during a 10-month product engagement.

• Built and maintained sales, inventory, and order-management dashboards for both products.
• Built frontend workflows for 100+ business-app integrations across the two platforms, including email and Google Sheets, with Shiprocket order/delivery integration and AI features in both products.
• Built a Next.js model-training UI for a major technology client, including reusable components, real-time training metrics, and code splitting.
• Led the frontend for the Kuberha AI website and admin dashboard using Next.js, Tailwind CSS, and shadcn/ui, with production role-based access.

Robotronix Engineering Tech · Indore, India

• Sole developer on Indore Carbonic, building and shipping a carbon-product ordering platform end to end with React.js and REST APIs.

Projects

EvalAgentFlow / Python · LangChain

An eight-agent pipeline that turns raw CSV evaluation data into benchmark reports, covering ingestion, cleaning, statistics, anomaly detection, insight extraction, and export. It blends human-eval rubrics with CodeBLEU, n-gram similarity, and dataflow-match metrics.

Cohart Study Guide / TypeScript · Next.js

An episode-based study platform covering React and Next.js architecture, Node.js and MongoDB backends, generative AI engineering, DSA, and system design. Built the viewer application and deployed it on Vercel. course.soujanyalab.in ↗

RideMate / MERN · WebSockets · Redis

MongoDB geospatial ride matching with live WebSocket updates, Redis caching, Node.js clustering, and a dedicated women-to-women safety mode. Source ↗

Skills

Languages

Python · JavaScript · TypeScript · C++ · SQL

AI & evaluation

Agent benchmarking · OpenClaw · WildClawBench · LLM-as-judge councils · Multi-agent orchestration · LangChain · RAG · Prompt engineering · CodeBLEU · Human-eval rubrics · NLP · Dataset quality

Web engineering

React · Next.js · Node.js · Express · FastAPI · Tailwind CSS · shadcn/ui · Redux Toolkit · TanStack Query · WebSockets

Data & infrastructure

AWS Bedrock · LiteLLM · Docker · Docker Compose · EC2 · CI/CD · MongoDB · MySQL · Redis · SQLite · pytest · REST APIs · Git · Git LFS

Education & certifications

Medi-Caps University

Certifications: AWS Academy Cloud Architecting · AWS Academy Cloud Foundations