PaperLens
Upload a research PDF, highlight passages, add notes, ask follow-up questions, and export an annotated PDF — no library or account required.…
● live systemTarun Sharma — AI/ML Consultant
Independent AI Consultant
From ambitionto productionintelligence
I partner with teams who need more than a pilot — who need AI systems that survive contact with reality. Evaluation suites that tell the truth. RAG pipelines that retrieve. Agentic workflows that deliver. And proof you can click, not promises you have to take on faith.
01
Measure
Rigorous evaluation before deployment — custom benchmarks, retrieval sweeps, and response-mode taxonomies that expose what models actually do.
02
Build
Agentic pipelines, RAG architectures, and knowledge systems engineered for production — not proof-of-concept theater.
03
Prove
Shipped demos, reproducible artifacts, and measurable outcomes clients can inspect, not slide decks they must trust.
Intelligence landscape
Frontier release radar
I maintain a live view of major LLM releases — benchmarks, context windows, and capability shifts — so client recommendations reflect what shipped this week, not last quarter.
- 159+Models tracked
- 9Labs covered
- ChatGPT · Nov 2022Since
Opens in a new tab — the tracker blocks embedding, but stays one click away.
AnthropicClaude Opus 4.xSWE-Bench leader
OpenAIGPT-5.xGPQA frontier
GoogleGemini 2.xMultimodal
DeepSeekR1 / V3Open-weight
Selected Work
Engagements that compound
Benchmarks, production systems, and deployable proofs — each built to demonstrate capability, not accumulate dust.
View →
Opportunity Radar
Daily business-problem discovery system: market signals → pain extraction → research evidence → ranked buildable AI/ML product briefs. Philo…
geo-kg
Builds a geometry knowledge graph from olympiad PDFs: problems, theorems, concepts, motifs, and proof tactics — with visualization and…
ECG-Bench
ECG = Epistemic · Coercion · GOAT — a 143-item benchmark that measures which response modes LLMs activate under fixed answer coercion (— one…
ModelBench
Personal model update tracker + benchmark suite for embeddings and rerankers on a private consultancy RAG eval set — with honest verdicts…
Quote Memory
Half-remember a line from a show? Quote Memory finds the episode and timestamp from fuzzy subtitle search across multiple series. What it…
● live systemHindi Jinnie
Interactive EN→HI translation demo with two model modes: a fast distilled LSTM and a more accurate Transformer trained from scratch on OPUS…
● live systemGet in touch
Ready to build something that lasts?
Open to consulting engagements, research collaborations, and transformation projects where measurement and delivery matter equally.
Contact