Senior AI Engineer · 7+ years building production LLM systems — RAG pipelines, agent orchestration, multi-provider LLM infrastructure, and voice AI on telephony. Based in India, working remote.
Available immediately for full-time remote senior AI engineer roles — based in India, working with global teams, comfortable with US/EU overlap.
Previously founding AI engineer at Kuration AI (Hong Kong) and sole AI hire reporting to the CTO at Schneider Electric / Luminous Power Technologies (India).
📫 irfan.ali@datacortex.in · LinkedIn · Résumé (PDF)
LLMs · RAG · Agents · Voice AI · Multi-provider orchestration
- RAG & retrieval — hybrid BM25 + dense, query rewriting, RRF fusion, evaluation
- Agents — multi-agent pipelines, structured extraction, tool use
- Voice AI — phone-based LLM agents on telephony, async webhooks, post-call analysis
- LLM infrastructure — multi-provider orchestration (GPT-4o, Claude, Gemini) with automatic fallback and cost controls
Python · FastAPI · LangChain · LangGraph · DSPy · LlamaIndex · pgvector · PostgreSQL · Bolna · Twilio · OpenAI · Anthropic · Groq
- Godam — a B2B inventory & billing SaaS for FMCG distribution in Nepal, live in production with a paying customer. Runs the client's daily billing, payments, credit ledgers, and field-sales logging. Next.js 15 · TypeScript · Supabase. Built and sold solo.
- Stacksift — a B2B domain product analyzer running a 5-stage LLM classification pipeline (DSPy + GPT-4.1) over crawled web data at ~$0.03 per analysis (~84% gross margin), with live Razorpay payments.
- Reflecta — a voice-first AI wellness app you can call on the phone. Phone check-ins → post-call analysis extracting 35+ wellness signals per call (burnout risk scoring, crisis-keyword detection) → personalised recommendations. FastAPI · Bolna telephony · Groq · Neon Postgres/pgvector · multi-provider fallback. Solo build, end-to-end.
| Library | Downloads | What it does |
|---|---|---|
| RAGNav |  |
RAG routing & retrieval — hybrid BM25 + dense with RRF fusion. R@3 = 0.956 on SQuAD, 131 tests |
| ragfallback |  |
Resilient retrieval — query rewriting, confidence scoring, fallback & retry |
| AgentEnsemble |  |
Multi-agent orchestration — ReAct, Swarm, Pipeline, Debate, WorkflowGraph |
| AgentCare |  |
Voice AI for healthcare — call intake, structured extraction, appointment orchestration |
| nepal-gov-agent |  |
Multilingual RAG over Nepal government documents, with a synthetic QA evaluation harness |
| AskPandas |  |
Natural-language queries on CSV via local LLMs — no API keys, no cloud |
| PyroChain |  |
Agentic feature engineering — PyTorch + LangChain agents for multimodal extraction |
| scrapeflow-py |  |
Playwright scraping with LLM extraction, hybrid selectors, anti-detection |
| lingo-nlp-toolkit |  |
Lightweight NLP utilities bridging classic pipelines and transformer workflows |
| toxic-comment-classifier |  |
Deep-learning toxicity detection with per-category scoring |
| socialmediaextractor |  |
Structured data extraction from social media profiles and posts |
| trustpilot-scraper |  |
Review collection and structuring from Trustpilot pages |
→ All libraries on my PyPI profile.
- Mental Health AI on MentalChat16K — BERT + neural networks on a cross-validation framework · IJAINN, Dec 2025 · DOI
- Neural-Symbolic Topic Evolution on Yelp Reviews — multi-aspect temporal topic modelling · IJAINN, Oct 2025 · DOI
Reliability over hype · Systems over scripts · Maintainability over short-term hacks.