Token Maxxer | AI Engineer
Vedika.io | Remote | Full-time
CLAUDE MASTER. TOKEN MAXXING. Engineering judgment required.
We’re building an AI-native company around APIs and personalized-advice software. We’re looking for an engineer who wants to understand the entire system: the data a model learns from, the evidence used to evaluate it, and the infrastructure that turns its output into a reliable product.
You’ll contribute to technical work across LLM training and fine-tuning, evaluation, inference and agent workflows. The goal is measurable improvement, not a convincing demo that fails outside its happy path.
WHAT YOU’LL WORK ON
- Training data: curate examples, define annotation criteria, identify duplicates and leakage, and construct useful train, validation and test splits.
- Model adaptation: help design fine-tuning experiments where appropriate; compare them against prompting and retrieval baselines before assuming training is necessary.
- Evaluation: build representative test sets, error taxonomies and regression checks. Assess task accuracy, instruction-following and unsupported claims separately.
- Inference: investigate latency, token usage and cost; test context selection, caching and model routing without silently sacrificing output quality.
- Agent systems: develop tool-calling, MCP and computer-use workflows with validation, bounded permissions, retries and observable failure paths.
- Product engineering: build APIs and integrations, write tests and deliver changes another engineer can reproduce, review and maintain.
MANDATORY WORKING KNOWLEDGE
- Python or TypeScript, REST APIs, SQL, Git and testing.
- Claude Code and Codex: practical workflows for investigating, implementing and verifying changes.
- Computer-use tools: browser or desktop workflows, outcome verification and approval boundaries.
- Slack and Jira/Atlassian: clear communication, documented decisions and accurate task/blocker tracking.
Familiarity with PyTorch, Hugging Face, parameter-efficient fine-tuning, Docker or cloud deployment is valuable. You are not expected to be an expert in every area. Strong fundamentals, careful experimentation and the ability to learn matter.
HOW WE WANT TO WORK
High ownership, fast feedback and substantive technical discussion. Question assumptions respectfully. Make uncertainty explicit. Share failures early. Leave evidence and documentation that help the next person move faster.
ENTRY ROUTES
Candidates with relevant experience will be considered directly for full-time employment. Candidates without relevant experience will start with a six-month internship, with a performance-based pre-placement offer (PPO) opportunity. A PPO is not automatic or guaranteed.
Candidates from Tier 1 institutions are preferred. Strong applicants from other institutions are welcome and should demonstrate relevant ability through their work.
APPLY WITH EVIDENCE
Share a project, repository or experiment. Explain your contribution, baseline, evaluation method, failure cases and what the results actually support.
Hiring urgently. Read the complete JD before applying. Apply only if you can meet the full-time commitment and demonstrate working knowledge of the mandatory tools. Include your relevant experience and earliest realistic joining date.