ASAIF ALI.

I’m Asaif Ali.

I build AI systems that hold up beyond the demo.

A little about meSee my work

Based in Chandigarh, India

I enjoy the space between experimentation and dependable software — trying things, learning fast, and making the next version better.

LinkedIn ↗GitHub ↗Email ↗
Asaif Ali
About me
A little more about the person behind the projects.

I like the part of engineering where an idea has to become real.

I’m an AI/ML engineer who enjoys the space between experimentation and dependable software. I like trying new models and frameworks, but I’m equally interested in what happens after the interesting demo: messy inputs, incomplete evidence, long-running jobs, failure paths, evaluation, and the details that make a system trustworthy.

Most of my work revolves around building useful AI products — from retrieval and document intelligence to agentic workflows and AI-assisted software engineering. I care about clear interfaces, measurable behavior, and systems that can be explained to the person using them.

This site is a record of that work, but more importantly, it is a place to get a sense of how I think and what I enjoy building.

01Curious first.

I like learning by building, testing ideas, and getting close to the real problem.

02Practical always.

A clever model is useful only when the surrounding system makes it dependable.

03Keep improving.

Ship something real, learn from it, then make the next version better.

Experience & focus

The work I do professionally — and the things I’ve learned building it.

My day job gives me real engineering problems to solve. Outside work, I take the ideas that stay with me and turn them into personal systems, experiments, and working products.

NOV 2024 — AUG 2026

Software Engineer · AI/ML

Bebo Technologies Pvt Ltd

I work on AI/ML systems that sit close to real product and engineering problems — building automation, retrieval and agentic workflows, and the infrastructure around them. A lot of the work is less about choosing a model and more about making the whole system behave well: connecting APIs and data sources, structuring context, validating outputs, handling failure paths, and creating interfaces that people can actually use.

Over time, that has led me deeper into agentic systems, RAG, multi-agent orchestration, application modernization, and AI-assisted engineering. I enjoy the point where a promising experiment becomes a dependable piece of software — and I’ve learned to think about evaluation, persistence, observability, deployment, and edge cases as part of the product rather than afterthoughts.

The personal projects on this site grow out of that same curiosity. They are where I get to ask, “What would this look like if I took the idea a little further?”

01Agentic systems

RAG, tool use, orchestration, memory, and multi-agent workflows around real application problems.

02Engineering around models

APIs, validation, retrieval, evaluation, persistence, deployment, and the pieces that make AI usable.

03From idea to software

Taking an experiment beyond the demo and thinking through quality, failure paths, and operations.

What I work with

The practical stack

I keep the toolkit flexible. The common thread is Python, modern LLM tooling, retrieval, APIs, and cloud infrastructure.

AI & LLMsGenerative AI · Agentic AI · OpenAI API · Gemini API
Agent systemsLangGraph · LangChain · Agno · CrewAI
Retrieval & searchOpenSearch · BM25 · Sparse retrieval · RRF · Jina · Tavily
Backend & dataPython · FastAPI · REST APIs · SQL · PostgreSQL · Redis
InfrastructureDocker · Supabase · Vercel · Render · AWS
Selected personal builds

A few things I’ve built outside the day job.

These are the projects where I give myself room to explore a problem, test an idea, and take it far enough to become a real working system.

01 / 05Featured build
Verified sparse-first RAG & research

EvidenceFlow

Sparse-first RAG and research with vectorless inverted-index retrieval, adaptive recovery, reranking, evidence verification, and fail-closed answers.

Current Scenario

RAG systems can retrieve plausible text while still producing unsupported claims, overconfident answers, or poor results when dense-vector retrieval is unavailable or poorly matched to the query.

What I built

A sparse-first retrieval pipeline uses OpenSearch's inverted index and BM25 as a vectorless foundation, with exact, phrase, fuzzy, metadata-aware, and optional neural-sparse retrieval. Adaptive query recovery, reciprocal-rank fusion, cross-encoder reranking, evidence tracking, and deterministic citation verification turn retrieved material into a controlled evidence set before answer generation.

LangGraphOpenSearchBM25RRFJina AI
Explore project GitHub
Built beyond the individual projects

AI Engineering Hub

One reusable engineering layer across five AI systems — with project adapters, task-specific evaluation, LLMOps telemetry, guardrails, human approval, regression gates, and n8n orchestration.

5 AI systemsTask-specific evalsLLMOps telemetryGuardrails + HITLCI regression gatesn8n orchestration
Explore the AI Engineering Hub
Get in touch

Interested in working together?

For a role, collaboration, or just a technical conversation, I’d be happy to hear from you.

EMAILEmail meLINKEDINConnect with me on LinkedInGITHUBSee what I’m building