Headline

AI Engineer @ Allyra.AI | Agentic AI • Multi-Agent Systems • RAG • MCP • Document Intelligence • FastAPI • LLM Applications

About

I am an AI Engineer focused on building production-grade AI systems that move beyond demos and into real-world workflows. Currently, I work at Allyra.AI, where I contribute to HireHub, an AI-powered recruitment orchestrator built on a multi-agent architecture. My work spans debugging and optimizing agent workflows, building LinkedIn profile enrichment systems, developing data normalization pipelines, and improving the reliability of AI-driven hiring processes. Previously, I worked with EaseMyTrip on agentic AI workflows using Model Context Protocol (MCP), function calling, and real-time travel data integrations. I also built hybrid retrieval systems combining OpenAI embeddings, BM25, and Reciprocal Rank Fusion (RRF) for large-scale travel knowledge bases. My experience includes multimodal document intelligence for healthcare applications, where I worked on OCR pipelines, vision-language models, semantic retrieval systems, and clinical document processing. I have also built data-intensive analytics pipelines and machine learning solutions across startup and contract environments. Key areas of interest: • Agentic AI & Multi-Agent Systems • MCP (Model Context Protocol) • Retrieval-Augmented Generation (RAG) • Document Intelligence & Multimodal AI • LLM Applications & Tool Calling • Semantic Search & Knowledge Systems • FastAPI & Production AI Infrastructure Always open to discussions around AI engineering, agentic workflows, retrieval systems, and emerging LLM technologies.

Location

Delhi

AI stack

codex

cursor

Work experience

Software Engineer (AI) at allyra.ai, May 2026 - Present · 4 mos, On-site, 1. Build HireHub, a live production hiring platform on Google ADK + FastAPI — backend microservices with a central orchestrator agent automating JD creation, resume matching, screening, scheduling, and evaluation. 2. Shipped a market-research feature for the JD agent that pulls live role-specific hiring trends (Gemini + Google Search) into JD generation — producing market-aligned descriptions instead of generic content, with fail-safe fallbacks and role-level caching. 3. Built a timed auto-send evaluation service that generates and emails candidate evaluations 45 minutes post-interview, with no-show handling and feature-flag-gated rollout. 4. Designed phase-aware agent orchestration using deterministic logic over prompt-only control, so agents take the right action per workflow stage — eliminating a recurring class of wrong-suggestion bugs. 5. Built a unified A2A integration layer routing chat, JD upload, and CV upload through a single message/send endpoint — with attachment handling, cross-worker session resolution, and idempotency to prevent duplicate pipeline runs. 6. Hardened production reliability — resolved state-consistency, distributed-workflow, and cross-service auth defects that surfaced only with real users (e.g. a candidate-identity bug leaking screening status across unrelated roles; orphan-candidate creation before email auth).

Software Engineer Trainee (AI) at EaseMyTrip.com, Mar 2026 - May 2026 · 3 mos, Gurugram, Haryana, India, • Built agentic AI workflows using MCP (Model Context Protocol) to integrate real-time flight pricing and travel data into AI-driven product features. • Designed multi-step tool-use pipelines with structured function calling, enabling AI agents to query live travel APIs and surface context-aware recommendations. • Implemented prompt grounding and output validation layers to ensure reliability of agentic responses in production travel use cases. • Built a hybrid semantic search engine over a 70-section travel knowledge base (Flight, Train, Bus, Hotel) using OpenAI embeddings + BM25 fused via Reciprocal Rank Fusion (RRF) with local .npz caching; validated end-to-end with 11 integration tests. • Built a data pipeline to transform raw chat logs into structured QA pairs for RAG, improving data usability and retrieval quality.

AI Engineer at Freelance, Apr 2025 - Oct 2025 · 7 mos, Remote, • Designed and developed a Healthcare RAG Assistant to query multi-format medical and regulatory documents (PDF, CSV, Excel, HTML, TXT). • Implemented FAISS + OpenAI embeddings for scalable retrieval and built a custom prompt framework to ensure compliance and factual accuracy. • Integrated an AI-powered PDF Image Interpreter (PyMuPDF, PIL, Gemini API) to extract and summarize images, automating image-to-text explanations for compliance and medical reviews. • Authored a detailed RFC outlining ingestion workflows, chunking strategies, metadata schema, semantic indexing, and retrieval pipelines, and benchmarked results showing that the RAG system reduced insight retrieval time from 45–90 min/report to under 10 seconds (~99% faster), improved accuracy from ~65–70% to over 90% (+25% gain), and cut analyst weekly effort from 10–15 hours to under 2 hours (-80% effort). • Proposed a low-temperature LLM response strategy to minimize hallucinations and enhance reliability in regulatory and clinical contexts.

Data Analyst at OrangeMetrics, Oct 2024 - Mar 2025 · 6 mos, Remote, 1. Utilized the GitHub API to extract and process over 50,000 commits from the Bitcoin repository, cleaning and organizing data with Pandas. 2. Categorized commits into Core, Wallet, and UX to generate actionable insights and analyzed contributions by language (C++ 40%, Python 20%, Shell 10%). 3. Produced more than ten Matplotlib visualizations, tracking monthly code churn, commit volume, category-wise contributions, language usage, top contributors, and cumulative trends.

Education

Bachelor's degree, Majors in Data Science, Indian Institute of Technology, Madras

Public links

github: https://github.com/HimanshuCraftsLab