Flagship systems get the visual weight. Supporting work stays concise. Each project points to the engineering decisions behind it.
Retrieval-first enterprise search built around page-aware ingestion, lexical + dense retrieval, fusion, reranking, evidence mapping and answer generation.
Evidence-oriented AI workflow for turning source material into structured claims with traceable support and explicit boundaries.
Semantic candidate–job matching and ranking with skill-gap signals.
Supporting application work demonstrating APIs, data flow and practical software engineering.
Technical credibility comes from showing architecture, measurements, boundaries and what remains unverified.
01 page-aware ingestion READY
02 lexical + dense retrieval FOUND
03 ranking / reranking ACTIVE
04 evidence mapping CHECK
05 answer boundary SET
Ingest → hybrid retrieval → fusion → reranking → evidence-aware generation. Retrieval is treated as a system, not a single vector search call.
Use verified measurements only. If a benchmark is small or internal, label it that way instead of turning it into marketing.
DSA is supporting evidence of problem-solving practice, not the identity of the portfolio.
VERIFY PROFILE ↗B.Tech in Artificial Intelligence & Data Science. Current portfolio record: CGPA 8.55/10.
Natural Language Processing and Business Intelligence & Analytics certifications completed in 2026.
Project status, metrics and credential claims should remain explicitly separated into verified, in-development and unverified areas.
Experiments are where questions become evidence before they become portfolio claims.
Compare retrieval quality, evidence coverage, answer faithfulness and failure cases instead of optimising only for a happy-path demo.
Inspect chunking, lexical retrieval, dense retrieval, fusion, ranking and evidence selection.
Test when an agent should retrieve, call a tool, ask for approval or stop. The interesting part is the boundary.
Compact experiments that explain one technique clearly. Less dashboard, more reproducible reasoning.
No fake percentages. These indicators describe demonstrated use represented in the portfolio.
I work across machine learning, information retrieval, LLM applications and backend engineering. My strongest projects are moving beyond model demos into retrieval evaluation, evidence grounding, semantic ranking and the engineering around them.
I care about understanding why a system behaves the way it does — not just making the demo pass. That means measuring quality, inspecting failure cases, keeping components modular and making trade-offs explicit.
I’m interested in AI/ML engineering, backend systems and software roles where there is something real to measure, debug and improve.