Hari Om Dwivedi — ML Engineer · AI Engineer · Software EngineerQuestions in. Answersout.
I'm Hari — a software engineer building production ML and LLM systems: RAG, semantic search, ranking and Spark-scale analytics. At Carnation Infotech, my NL→SQL platform cut analytics turnaround from 48 hours to under a minute.
The console is a real retrieval engine indexed on this page — BM25, an interpretable 16-d concept space and MMR re-ranking, all running in your browser.
ask-hari · hybrid retrieval · 62 passages
runs in your browser
Ask anything about Hari’s work.
Answers are retrieved from this page and cited — by a small search engine running in your browser. No LLM, no API calls, no tracking.
tokenize→
expand→
embed→
retrieve→
rerank→
cite
Try
Analytics turnaround
48h → <1 min
Conversational BI · Carnation Infotech
Lead-to-client conversion
+25%
Embedding-based ranking · Propques
Shopify bounty
$10,000
Two production Shop Minis
Students mentored
700+
President, CSI Student Chapter
01Featured work · Carnation Infotech
Talk to your data. From a 48-hour queue to answers in under a minute.
An enterprise conversational-AI platform for the Business Intelligence team: stakeholders ask questions of company-wide big data in plain English, and get validated charts, tables and insights back.
The problem
Business questions had to wait on analysts to write and run queries against company-scale data. Analytics requests took around 48 hours to come back.
What I built
translate()An LLM-powered layer that turns stakeholder questions into optimized PySpark SQL.
execute()Secure query execution on Apache Spark clusters, with automated aggregation, validation and post-processing.
explain()Summarization and visualization pipelines that return charts, tables and plain-English insights.
protect()Caching, query optimization and guardrails for performance, correctness and cost efficiency.
govern()Access control and data governance, enforced with data-engineering, platform and leadership teams.
Before
48 h
After
< 1 min
Stack
Python
LLMs
Prompt engineering
PySpark
Spark SQL
Apache Spark
Caching
Guardrails
Data visualization
RBAC
Stakeholder asks
“Which 5 regions grew revenue fastest last quarter?”
waitingt = 0 ms
SpanTimeline05 s10 s15 s20 sTime
auth.check_access
guardrails.input
cache.lookup
llm.translate
sql.validate_optimize
spark.execute
results.validate
llm.summarize
viz.render
SparkLLMPlatform cache hit blocked
spark.execute · Spark · 18.6 s
Distributed query on the Spark cluster.
Revenue growth, last quarter vs previous (%)
North+18.4%
West+12.9%
South+9.7%
Central+6.1%
East+3.2%
Illustrative trace — synthetic schema, data and timings. The production system is internal to Carnation Infotech.
Analytics turnaround, drawn to scale1 : 2,880
Before
~48 hours
After
< 1 minute
At true scale the “after” bar is 1/2,880th the length of the “before” bar — about a third of a pixel here, so it’s drawn at the 1 px minimum with a marker.
02Experience
Where I’ve shipped.
Software engineering across LLM platforms, search and recommendations, and SaaS backends — always measured by what reached production.
Aug 2025 — Present
Carnation Infotech
Converted from intern to SDE I
Software Development Engineer I
Oct 2025 — Present● now
Conversational AI for Business Intelligence
48h → <1 minanalytics turnaround
Designed and built an enterprise-scale conversational AI platform for the Business Intelligence team, enabling natural-language interaction with company-wide big data.
Architected an LLM-powered query translation layer that converts stakeholder questions into optimized PySpark SQL queries.
Integrated secure query execution on Apache Spark clusters with automated aggregation, validation and post-processing of results.
Developed summarization and visualization pipelines that return charts, tables and insights in a user-friendly format.
Implemented caching, query optimization and guardrails to ensure performance, correctness and cost efficiency.
Reduced analytics request turnaround from 48 hours to under 1 minute, dramatically improving stakeholder productivity.
Collaborated with data engineering, platform and leadership teams to enforce access control, data governance and scalability.
Attention lets each token weigh every other token when building its representation…12
LLM runs locally
Whisper
Local GenAI assistant on WhatsApp
A privacy-first assistant that answers questions over WhatsApp with retrieval-augmented generation — the LLM runs locally, not behind a third-party API.
Built a privacy-first local LLM assistant providing real-time question answering via WhatsApp.
Implemented retrieval-augmented generation using embeddings and a vector database (ChromaDB).
Designed ingestion, chunking, retrieval and prompt-optimization pipelines for accuracy and low latency.
Machine learning, LLMs and retrieval, data engineering, backend and cloud. Each skill is embedded from where I’ve actually used it, so its nearest neighbours are computed — not curated. Pick any skill.
Dot size = roles & projects that used the skill · lines = 4 nearest neighbours by cosine similarity · grouped by domain
Product-driven software engineer building scalable backends, AI platforms and data-intensive applications — documented in the format ML teams use for models.
Production LLM platforms, RAG & semantic search, PySpark at company scale, ranking & recommendations
Context window
The full lifecycle: system design → implementation → deployment → monitoring → iteration
Intended use
ML Engineer, AI Engineer, LLM Engineer and SDE roles on teams shipping AI products
Out-of-scope use
Demos that never leave the notebook
Training regime
Four national hackathon awards, a $10K Shopify bounty and real production deadlines
Known limitations
Turns portfolios into retrieval systems. You may have noticed.
Evaluation results
Benchmark and source
Score
Analytics turnaroundCarnation Infotech · BI
48 h → <1 min
Lead-to-client conversionPropques · PropSaaS
+25%
Shop Minis bountyShopify
$10,000
Students mentoredCSI Student Chapter
700+
National hackathon awardsAKTU · IIIT-L · HackCBS · UP Police
4
Citation
@misc{dwivedi2026,
author = {Dwivedi, Hari Om},
title = {Hari Om Dwivedi: ML \& AI Engineer},
year = {2026},
howpublished = {\url{https://www.hariom.work/}},
note = {Open to ML Engineer, AI Engineer and SDE roles}
}
Education & leadership
B.Tech, Computer Science & Engineering
SRMCEM (affiliated with AKTU), Lucknow · 2021 — 2025 · First Division
President, CSI Student Chapter — led technical initiatives, organized large-scale hackathons and mentored 700+ students.
Recognized for leadership, system design and innovation through multiple national-level hackathon awards.
06Recognition
The leaderboard.
National-level hackathon results, a Shopify bounty, and the certifications behind the fundamentals.
I’m open to Machine Learning Engineer, AI Engineer and Software Development Engineer roles — especially on teams putting LLMs, retrieval and data platforms into production. Email is the fastest way to reach me.