Applied AI Engineer · IST (UTC+5:30) · Open to remote roles
AI Engineer with 2+ years building and deploying production AI/ML systems — LLM fine-tuning, RAG, agentic pipelines, and model evaluation. Track record optimizing model quality, latency, and inference cost at scale, backed by open-source contributions to major AI/ML repositories. Focused on turning research techniques into reliable, reusable AI infrastructure.
- AI/ML: LLMs, Transformers, RAG, Fine-Tuning, LoRA, Quantization, Embeddings, NLP, Computer Vision, VLMs, Prompt Engineering, AI Agents
- Languages: Python, TypeScript, Go
- Cloud & DevOps: AWS, Kubernetes, Docker, Terraform, GitHub Actions, Linux
- Frameworks & Tools: PostgreSQL, SQLAlchemy, Redis, Git
- Web3/Blockchain: Solidity, Foundry, Hardhat, EIP-712, UUPS Upgradeable Proxies, Gnosis Safe, The Graph, Cross-Chain Messaging (CCTP), Gas Optimization
Applied AI Engineer, Syvora — July 2025 – Present
- Developed document-processing workflows combining OCR, VLM inference, validation, and structured extraction, reducing manual processing time by 80%.
- Built AI services integrating LLMs, RAG, PostgreSQL, Redis, and REST APIs, handling 100K+ requests/day across production workflows.
- Implemented model routing, caching, batching, and provider fallbacks, reducing AI infrastructure costs by 61% while maintaining 95%+ task success.
- Deployed AI workloads on Kubernetes, implementing GPU scheduling, autoscaling, health checks, rolling deployments, and model version management.
Protocol Engineer, NDA — Jun 2024 – May 2025
- Implemented defense-in-depth for LLM applications using prompt-injection detection, retrieval isolation, tool authorization, output validation, and policy checks.
- Architected Polymarket-style prediction exchange contracts with deterministic order matching, invariant-safe execution, and settlement state machines sustaining 10k+ daily on-chain trades in production.
- Authored 300+ Foundry fuzz and invariant tests covering adversarial execution paths including duplicate messages and partial settlement, achieving 97% branch coverage on core protocol contracts.
- Built cross-chain USDC settlement infrastructure processing 500k+ messages with domain-separated replay protection and invariant-safe execution, sustaining zero failed settlements across all integrated chains.
Pull requests across the AI/ML and infrastructure projects below (merged and currently open; counts update automatically).
| Repository | Merged | Open | Focus |
|---|---|---|---|
| temporalio/ai-integrations | 7 | 1 | OpenAI Agents plugin: tool metadata, tool argument errors, RunState resume; Gemini permanent-error handling |
| roboflow/supervision | 4 | 0 | Mask-hole dataset export, CompactMask.resize parity, CSV export of array data |
| ComposioHQ/composio | 3 | 0 | Anthropic prompt caching and OpenAI Agents strict mode (Python), cache-breakpoint fix (TypeScript) |
| lancedb/lancedb | 3 | 0 | Embedding retry/backoff on permanent errors, optimize() keeping tagged versions |
| docling-project/docling | 2 | 2 | Office backends: DOCX/ODF footnotes, PPTX formatting and hyperlinks |
| mlflow/mlflow | 2 | 2 | Experiment and run restore, prompt search ordering, GCS/Azure presigned URLs |
| truera/trulens | 2 | 1 | Bedrock and LangChain provider fixes, Cursor token usage |
| getlago/lago-api | 2 | 0 | Coupon and add-on lifecycle webhooks |
| collabora/WhisperLive | 1 | 1 | OpenVINO backend: prompts, hotwords, diarization |
| temporalio/sdk-python | 1 | 0 | Workflow update validator checks, Annotated parameter hints |
| weaviate/weaviate | 0 | 4 | Backup API, tenant filtering, reranker config |
| apache/airflow | 0 | 2 | Listener hooks for task retries and DAG pause state |
| celery/celery | 0 | 2 | Chord errbacks given as a single signature |
| etcd-io/etcd | 0 | 2 | etcdctl watch --from-key, concurrency.ResumeMutex |
| promptfoo/promptfoo | 0 | 2 | exec: provider argument quoting, unquoted YAML dates in config and test files |
| qdrant/qdrant | 0 | 2 | gRPC metrics whitelist, public gRPC Quotas service |
Auto-updated by a scheduled GitHub Action — see workflow.
- [lancedb/lancedb] fix(python)!: raise on bool(Expr) instead of silently dropping predicates — merged
- [temporalio/ai-integrations] Let Strands recover from model context window overflow in TemporalAgent — merged
- [temporalio/ai-integrations] Carry binary file content across the Gemini file upload activities — merged
- [celery/celery] Fix disable_events on a worker started with -E — open
- [mlflow/mlflow] Preserve NaN logged model metrics in
SqlAlchemyStore— closed - [roboflow/supervision] fix(vlm): crop Gemini 2.5/3.5 masks to the image when box_2d leaves 0-1000 — merged
Last updated 2026-10-27 11:29 UTC.
The Curious Crunch — an ongoing technical blog series reverse-engineering the system infrastructure behind Web3, AI, and DevOps.
Acropolis Institute of Technology & Research — B.Tech, Computer Science